DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed.
Claim Objections
Claim 12 is objected to because of the following informalities: Claim 12 recites “replay” in lines 4 and 6 of the claim, which is a spelling error. Examiner suggests revising the claim to recite “reply.” Appropriate correction is required.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-4, 7-12, 14, 17-19, 22, and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Saini et al. (US Publication No. 2025/0061136) in view of Mahfuz et al. (US Publication No. 2024/0134908).
As to claim 1, Saini teaches a sentence generation method, comprising:
obtaining a query sentence [input query], wherein the query sentence has at least one candidate description object [query embedding] (see e.g., [0079] for the computing system 102 receiving the input query 104 and the computing system 102 encoding the input query 104 into a query embedding EQ (e.g., the query embedding 228). The computing system obtains an input query, which has a query embedding.);
performing semantic recognition on the query sentence to obtain first semantic information [metadata embedding representing user location] corresponding to the query sentence (see e.g., [0028] for a first metadata model 202 (M1) transforming the input query 104 to a first lookup vector 204, and then using the first lookup vector 204 to identify a first metadata embedding EM1 in a first metadata vector space 206 and each of the above-identified metadata embeddings being associated with a particular metadata item, [0030] for embeddings in each vector space having a relationship to each other and in particular, the distance between any pair of vectors in a single vector space reflecting the semantic similarity between the concepts associated with the two vectors, [0036] for different applications using metadata models that produce different kinds of metadata, and [0041] for metadata model receiving location information that identifies the location associated with the user and the metadata model leveraging location information in various ways, such as by adding text to the input query 104 that identifies the location of the user, and then transforming the thus-augmented query into a query embedding. The computing system obtains a metadata embedding representing location of the querying user by recognizing that the embedding is semantically similar to the input query augmented with location information.);
inputting the at least one candidate description object and the first semantic information into a large model [generative model], to identify a target description object [keyword] from the at least one candidate description object based on the large model (see e.g., [0024] for a query-augmenting component 110 combining the query embedding with the metadata embedding(s), to produce an augmented embedding EQ+ and a retrieval component 112 using the augmented embedding to identify one or more target items and [0049] for a retrieval component 602 that represents the retrieval component 112 of FIG. 1, the retrieval component 602 using a machine-trained generative model 604 to generate one or more target items based on the augmented embedding EQ+, for example, the machine-trained generative model 604 treating the augmented embedding as a prompt, and using a language model of any kind to autoregressively transform the prompt into at least one item, token by token, and for instance, the retrieval component 602 transforming the augmented embedding into a keyword (which includes one or more words), and then finds an ad associated with the generated keyword. The query embedding and the metadata embedding are input into a generative model to identify a keyword from the input query based on the generative model.);
selecting target service data [ad] from at least one piece of service data [target data] corresponding to the target description object (see e.g., [0049] for a retrieval component 602 that represents the retrieval component 112 of FIG. 1, the retrieval component 602 using a machine-trained generative model 604 to generate one or more target items based on the augmented embedding EQ+, for example, the machine-trained generative model 604 treating the augmented embedding as a prompt, and using a language model of any kind to autoregressively transform the prompt into at least one item, token by token, and for instance, the retrieval component 602 transforming the augmented embedding into a keyword (which includes one or more words), and then finding an ad associated with the generated keyword. An ad is selected from the target data corresponding to the keyword.); and
generating a reply sentence [output information] corresponding to the query sentence according to the target service data (see e.g., [0024] for an output-generating component 116 providing output information 118 that includes each identified target item and the output-generating component 116 producing output information 118 which contains an ad that matches the input query 104. Output information corresponding to the input query is generated according to the ad. Output information may encompass a sentence within the ad in reply to the input query.).
Saini does not specifically disclose performing semantic recognition on the query sentence to obtain second semantic information corresponding to the query sentence, wherein categories of the first semantic information and the second semantic information are different; and selecting target service data from at least one piece of service data corresponding to the target description object based on the second semantic information. However, Mahfuz teaches
performing semantic recognition on the query sentence [query] to obtain second semantic information [context term indicating a range of time stamps] corresponding to the query sentence, wherein categories of the first semantic information and the second semantic information are different (see e.g., [0057] for the query 120 including “where is that video of the bell ringing multiple times that I took last week?” and “last week” being a context term indicating a range of time stamps when the target media file was created and [0069] for the natural language processor 302 being configured to process the text query 202 to identify the portion of the text query 202 that corresponds to the target sound description 304 and the portion of the text query 202 that corresponds to the context term(s) 306. The natural language processor performs semantic recognition on the query to obtain a context term indicating a range of time stamps, which corresponds to the query. A range of time stamps is a different category from location of the querying user.); and
selecting target service data [highest-ranked search results] from at least one piece of service data [media file] corresponding to the target description object [sound caption] based on the second semantic information (see e.g., [0034] for if the query includes natural-language text, at least a portion of the text being used to generate a query caption embedding (e.g., a sentence embedding) and in this example, the query caption embedding being compared, in a caption embedding space, to caption embeddings associated with the media files to determine a similarity metric, and the search results being determined based on the similarity metric, [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156, [0066] for the search results 124 being limited to include only a specific number of most similar results, [0070] for filter criteria being based on the context term(s) 306 and filter criteria including a time criterion, and [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124 and for example, in some such implementations, the comparator 140 determining similarity metrics for caption embeddings 160 associated with metadata 154 that failed to satisfy the filter criteria; however, such caption embeddings being unfavorably weighted to appear lower in the ranked search results 124 than they would had their metadata 154 satisfied the filter criteria. The highest-ranked search results are selected from at least one media file corresponding to the sound caption based on the context term indicating a range of time stamps).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini to perform semantic recognition on the query sentence to obtain second semantic information corresponding to the query sentence, wherein categories of the first semantic information and the second semantic information are different; and select target service data from at least one piece of service data corresponding to the target description object based on the second semantic information, as taught by Mahfuz, for the benefit of determining results relevant to a time range (see e.g., Mahfuz, [0057]).
As to claim 2, the limitations of parent claim 1 have been discussed above. Saini teaches
wherein the first semantic information is semantic information of a geographic area dimension (see e.g., [0041] for metadata model receiving location information that identifies the location associated with the user and the metadata model leveraging location information in various ways, such as by adding text to the input query 104 that identifies the location of the user, and then transforming the thus-augmented query into a query embedding. Location of a querying user is a geographic area dimension.).
Saini does not specifically disclose wherein the second semantic information is semantic information of a statistical frequency dimension related to the candidate description object, and/or semantic information of a statistical time dimension related to the candidate description object. However, Mahfuz teaches
wherein the second semantic information is semantic information of a statistical frequency dimension related to the candidate description object, and/or semantic information of a statistical time dimension related to the candidate description object [caption] (see e.g., [0057] for the query 120 including “where is that video of the bell ringing multiple times that I took last week?” and “last week” being a context term indicating a range of time stamps when the target media file was created, [0069] for the natural language processor 302 being configured to process the text query 202 to identify the portion of the text query 202 that corresponds to the target sound description 304 and the portion of the text query 202 that corresponds to the context term(s) 306, and [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124 and for example, in some such implementations, the comparator 140 determining similarity metrics for caption embeddings 160 associated with metadata 154 that failed to satisfy the filter criteria; however, such caption embeddings being unfavorably weighted to appear lower in the ranked search results 124 than they would had their metadata 154 satisfied the filter criteria. A range of timestamps is a statistical time dimension related to the caption corresponding to the caption embedding).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the second semantic information is semantic information of a statistical frequency dimension related to the candidate description object, and/or semantic information of a statistical time dimension related to the candidate description object, as taught by Mahfuz, for the benefit of determining results relevant to a time range (see e.g., Mahfuz, [0057]).
As to claim 3, the limitations of parent claim 1 have been discussed above. Saini does not specifically disclose wherein the at least one candidate description object is determined based on a following manner: obtaining one or more first semantic vectors and a first keyword corresponding to the query sentence; determining first description objects matching the one or more first semantic vectors and second description objects matching the first keyword from a preset retrieval library; and determining the candidate description object according to the first description objects and the second description objects. However, Mahfuz teaches wherein the at least one candidate description object is determined based on a following manner:
obtaining one or more first semantic vectors [query caption embedding] and a first keyword [word from target sound description] corresponding to the query sentence (see e.g., [0083] for the natural language processor 302 being operable to determine the target sound description 304 from text of the query 120 and to provide the target sound description 304 to the tag embedding generator 408, the caption embedding generator 242, or both, in this example, the caption embedding generator 242 generating the query caption embedding 210 based on the target sound description 304, and in some implementations, the tag embedding generator 408 generating the query tag embedding(s) 410 based on the target sound description 304, and the caption embedding generator 242 generating the query caption embedding 210 based on the query tag embedding(s). A query caption embedding and a word from the target sound description corresponding to the query are obtained.);
determining first description objects [captions] matching the one or more first semantic vectors and second description objects [sound tags] matching the first keyword from a preset retrieval library [file repository] (see e.g., [0083] for the comparator 140 determining similarity metrics by comparing the query caption embedding 210 and the caption embeddings 162, [0084] for in implementations in which the target sound description 304 is processed by the tag embedding generator 408 to generate the query tag embedding(s) 410, the comparator 140 also determining similarity metrics by comparing the query tag embedding(s) 410 and the tag embeddings 514 in a tag embedding space, [0089] for search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and FIG. 5 for the file repository including captions and sound tags. Comparison is used to determine captions and sound tags corresponding to search results. The captions match the query caption embedding and the sound tags match the target sound description.); and
determining the candidate description object [highly-ranked caption or sound tag] according to the first description objects and the second description objects (see e.g., [0089] for when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, a second weight being applied to similarity values based on comparisons of the audio embeddings 530 to the query audio embedding(s) 406, and a third weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. A highly ranked caption or sound tag is determined according to the caption and sound tags.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the at least one candidate description object is determined based on a following manner: obtaining one or more first semantic vectors and a first keyword corresponding to the query sentence; determining first description objects matching the one or more first semantic vectors and second description objects matching the first keyword from a preset retrieval library; and determining the candidate description object according to the first description objects and the second description objects, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 4, the limitations of parent claims 1 and 3 have been discussed above. Saini does not specifically disclose wherein the preset retrieval library is constructed based on a following manner: obtaining a service database, wherein the service database comprises multiple pieces of initial service data; determining service description objects of the initial service data, and data record types and data record frequencies corresponding to the service description objects; performing aggregation processing on the multiple pieces of initial service data based on the service description objects, the data record types, and the data recording frequencies, to obtain aggregated service data, wherein the aggregated service data comprises the at least one piece of service data corresponding to the target description object; performing vectorization processing on the service description objects based on a semantic vector model, to obtain a second semantic vector; and constructing the preset retrieval library based on the aggregated service data and the second semantic vector. However, Mahfuz teaches wherein the preset retrieval library is constructed based on a following manner:
obtaining a service database [set of media files], wherein the service database comprises multiple pieces of initial service data [media files] (see e.g., [0044] for the system 100 including a device 102 that includes one or more processors 190 and memory 192 and the memory 192 including a file repository 150 storing media files 152, [0080] for the media search engine 130 searching the file repository 150 based on the query 120, and [0081] for the file repository 150 including the set of media files 252. The device obtains a set of media files, which includes multiple media files.);
determining service description objects [captions] of the initial service data, and data record types [file types] and data record frequencies [sound frequency] corresponding to the service description objects (see e.g., [0057] for the query 120 including “where is that video of the bell ringing multiple times that I took last week?” and the term “video” being a context term indicating a file type of a target media file, [0070] for filter criteria being based on the context term(s) 306 and filter criteria including a target file type, [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124, and [0075] for the sound captioning engine 146 being configured to generate sound captions descriptive of sounds in the media files 152. Captions, which may include the fact that a sound occurred multiple times, are determined by the sound captioning engine. A file type corresponding to the caption is determined for weighting.);
performing aggregation processing on the multiple pieces of initial service data based on the service description objects, the data record types, and the data recording frequencies, to obtain aggregated service data, wherein the aggregated service data comprises the at least one piece of service data corresponding to the target description object (see e.g., [0081] and FIG. 5 for the file repository 150 including the set of media files 252, and each media file 152 being associated with metadata 154 and sound caption(s) 156A. The device aggregates media files with metadata, which includes file type, and captions, which includes sound frequencies. The aggregated media files include at least one media file corresponding to the target sound caption.);
performing vectorization processing [embedding] on the service description objects based on a semantic vector model [machine-learning model corresponding to caption embedding generator], to obtain a second semantic vector [caption embedding] (see e.g., [0051] for the sound captioning engine 146 including one or more machine-learning models and for example, the sound captioning engine 146 including one or more embedding generators 142, each of which corresponds to or includes at least one trained machine-learning model, [0052] for the sound captioning engine 146 being operable to generate embeddings 158 representing various data stored in the file repository 150 and for example, when a new media file 152 is added to the file repository 150, the sound captioning engine 146 generating a caption embedding 160 representing a sound caption 156 of the new media file 152, [0054] for each of the caption embeddings 160 being viewed as a vector designating a particular location in a caption embedding space. Embedding is performed on captions based on a machine-learning model corresponding to caption embedding generator to obtain a caption embedding, which is a vector.); and
constructing the preset retrieval library based on the aggregated service data and the second semantic vector (see e.g., [0081] and FIG. 5 for the file repository 150 including the set of media files 252, and each media file 152 being associated with metadata 154, sound caption(s) 156A, and corresponding caption embeddings 162.The file repository is constructed from media files with metadata, which includes file type, captions, which includes sound frequencies, and caption embeddings.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the preset retrieval library is constructed based on a following manner: obtaining a service database, wherein the service database comprises multiple pieces of initial service data; determining service description objects of the initial service data, and data record types and data record frequencies corresponding to the service description objects; performing aggregation processing on the multiple pieces of initial service data based on the service description objects, the data record types, and the data recording frequencies, to obtain aggregated service data, wherein the aggregated service data comprises the at least one piece of service data corresponding to the target description object; performing vectorization processing on the service description objects based on a semantic vector model, to obtain a second semantic vector; and constructing the preset retrieval library based on the aggregated service data and the second semantic vector, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 7, the limitations of parent claims 1 and 3 have been discussed above. Saini does not specifically disclose wherein the preset retrieval library comprises: multiple second semantic vectors, wherein determining the first description objects matching the one or more first semantic vectors from the preset retrieval library comprises: determining an initial similarity between a first semantic vector and each second semantic vector in the preset retrieval library; ranking the multiple second semantic vectors according to initial similarities; constructing a candidate vector set corresponding to the first semantic vector according to a first number of second semantic vectors ranked in the front; and determining the first description objects according to one or more candidate vector sets and a number of the one or more candidate vector sets. However, Mahfuz teaches wherein the preset retrieval library comprises:
multiple second semantic vectors (see e.g., [0081] and FIG. 5 for the file repository 150 including caption embeddings 162), wherein determining the first description objects matching the one or more first semantic vectors from the preset retrieval library comprises:
determining an initial similarity [similarity metric] between a first semantic vector and each second semantic vector in the preset retrieval library (see e.g., [0064] for the comparator 140 being configured to determine the similarity metric for each caption embedding 160 based on a distance 214, in the caption embedding space 220, between the caption embedding 160 and the query caption embedding 210 and the distance 214 being determined, for example, as a cosine distance, a Euclidean distance, or based on some other distance measure. A similarity metric is determined between a query caption embedding and each caption embedding in the file repository.);
ranking the multiple second semantic vectors according to initial similarities (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156, the search results 124 including a rank sorted list of results, and in such implementations, the search results 124 being sorted based on their respective similarity metrics. The multiple caption embeddings are ranked according to similarity metrics.);
constructing a candidate vector set [top-ranked caption embeddings] corresponding to the first semantic vector according to a first number of second semantic vectors ranked in the front (see e.g., [0066] for the search results 124 being limited to include only a specific number of most similar results. The top-ranked caption embeddings corresponding to the query caption embedding are determined.); and
determining the first description objects according to one or more candidate vector sets [top-ranked caption embeddings or top-ranked caption embeddings and top-ranked tag embeddings] and a number of the one or more candidate vector sets (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156 and the search results 124 identifying one or more media files 152 (or portions of media files 152, such as a portion of a media file 152 associated with a particular sound caption 156) associated with a set of the caption embeddings 160 that are most similar, among the set of media files 252, to the query caption embedding 210 and [0089] for the query 120 being used to generate multiple types of embeddings, which are compared to corresponding embeddings of the file repository 150 and for example, a query caption embedding 210 being compared to the caption embeddings 162 and query tag embedding(s) 410 being compared to tag embeddings 514 associated with the media files 152, when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, and a second weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. Top-ranked captions are determined based on top-ranked caption embeddings. Top-ranked captions are also determined based on whether only caption embeddings are being considered (one set of embeddings) or whether both caption embeddings and tag embeddings are being considered (two sets of embeddings).).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the preset retrieval library comprises: multiple second semantic vectors, wherein determining the first description objects matching the one or more first semantic vectors from the preset retrieval library comprises: determining an initial similarity between a first semantic vector and each second semantic vector in the preset retrieval library; ranking the multiple second semantic vectors according to initial similarities; constructing a candidate vector set corresponding to the first semantic vector according to a first number of second semantic vectors ranked in the front; and determining the first description objects according to one or more candidate vector sets and a number of the one or more candidate vector sets, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 8, the limitations of parent claims 1, 3, and 7 have been discussed above. Saini does not specifically disclose wherein there is one candidate vector set, and determining the first description objects according to the one or more candidate vector sets and the number of the one or more candidate vector sets comprises: performing weighting processing on the initial similarity of each second semantic vector according to a corresponding ranking order of the second semantic vector in the candidate vector set to obtain a target similarity corresponding to the second semantic vector; and determining the first description objects according to the target similarity of each second semantic vector. However, Mahfuz teaches wherein there is one candidate vector set [top-ranked caption embeddings], and determining the first description objects according to the one or more candidate vector sets and the number of the one or more candidate vector sets comprises:
performing weighting processing on the initial similarity of each second semantic vector according to a corresponding ranking order of the second semantic vector in the candidate vector set to obtain a target similarity [weighted similarity metric] corresponding to the second semantic vector (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156, [0066] for the search results 124 being limited to include only a specific number of most similar results, [0070] for filter criteria being based on the context term(s) 306 and filter criteria including a time criterion, and [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124 and for example, in some such implementations, the comparator 140 determining similarity metrics for caption embeddings 160 associated with metadata 154 that failed to satisfy the filter criteria; however, such caption embeddings being unfavorably weighted to appear lower in the ranked search results 124 than they would had their metadata 154 satisfied the filter criteria. Weighting processing is performed on the caption embedding to obtain a weighted similarity metric corresponding to the caption embedding. The weighting processing depends upon both the weight and the ranking order of the caption embedding in the top-ranked caption embedding set.); and
determining the first description objects according to the target similarity of each second semantic vector (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156 and the search results 124 identifying one or more media files 152 (or portions of media files 152, such as a portion of a media file 152 associated with a particular sound caption 156) associated with a set of the caption embeddings 160 that are most similar, among the set of media files 252, to the query caption embedding 210 and [0066] for the search results 124 being limited to include only a specific number of most similar results. Top-ranked captions are determined according to the weighted similarity of each caption embedding.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein there is one candidate vector set, and determining the first description objects according to the one or more candidate vector sets and the number of the one or more candidate vector sets comprises: performing weighting processing on the initial similarity of each second semantic vector according to a corresponding ranking order of the second semantic vector in the candidate vector set to obtain a target similarity corresponding to the second semantic vector; and determining the first description objects according to the target similarity of each second semantic vector, as taught by Mahfuz, for the benefit of determining results relevant to a time range (see e.g., Mahfuz, [0057]).
As to claim 9, the limitations of parent claims 1, 3, and 7 have been discussed above. Saini does not specifically disclose wherein there are multiple candidate vector sets, and different candidate vector sets are respectively obtained by performing vectorization processing on the query sentence based on different semantic vector models, wherein determining the first description objects according to the one or more candidate vector sets and the number of the one or more candidate vector sets comprises: performing weighting processing on the initial similarity of each second semantic vector according to a corresponding ranking order of the second semantic vector in each candidate vector set, to obtain a reference similarity corresponding to the second semantic vector; determining a sum of multiple reference similarities corresponding to the second semantic vector as a target similarity of the second semantic vector; and determining the first description objects according to the target similarity of each second semantic vector. However, Mahfuz teaches wherein there are multiple candidate vector sets [top-ranked caption embeddings and top-ranked tag embeddings], and different candidate vector sets are respectively obtained by performing vectorization processing [embedding] on the query sentence based on different semantic vector models [machine-learning model corresponding to caption embedding generator and machine-learning model corresponding to tag embedding generator] (see e.g., [0051] for the sound captioning engine 146 including one or more machine-learning models and for example, the sound captioning engine 146 including one or more embedding generators 142, each of which corresponds to or includes at least one trained machine-learning model, [0054] for each embedding of a particular type being viewed as a vector designating a point in an embedding space associated with that type of embedding, to illustrate, each of the caption embeddings 160 being viewed as a vector designating a particular location in a caption embedding space, and each of the tag embeddings being viewed as a vector designating a particular location in a tag embedding space, in this example, the comparator 140 determining similarity between a query embedding and an embedding of the embeddings 158 based on a metric (e.g., a similarity metric) associated with relative positions, in the appropriate embedding space, of the two embeddings, and [0083] for the natural language processor 302 being operable to determine the target sound description 304 from text of the query 120 and to provide the target sound description 304 to the tag embedding generator 408, the caption embedding generator 242, or both. Top-ranked caption embeddings, which are vectors, are obtained by embedding the query based on a machine learning model corresponding to caption embedding generator. Top-ranked tag embeddings, which are vectors, are obtained by embedding the query based on a machine learning model corresponding to tag embedding generator.), wherein determining the first description objects according to the one or more candidate vector sets and the number of the one or more candidate vector sets comprises:
performing weighting processing on the initial similarity of each second semantic vector according to a corresponding ranking order of the second semantic vector in each candidate vector set, to obtain a reference similarity [weighted similarity metric] corresponding to the second semantic vector (see e.g., [0089] for the query 120 being used to generate multiple types of embeddings, which are compared to corresponding embeddings of the file repository 150 and for example, a query caption embedding 210 being compared to the caption embeddings 162 and query tag embedding(s) 410 being compared to tag embeddings 514 associated with the media files 152, when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, and a second weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. Weighting processing is performed on the similarity metric of each embedding to obtain a weighted similarity metric corresponding to the embedding. The weighting processing depends upon both the weight and the ranking order of the embedding in each top-ranked embedding set.);
determining a sum of multiple reference similarities corresponding to the second semantic vector as a target similarity of the second semantic vector (see e.g., [0084] for the query tag embedding(s) 410 include more than one tag embedding for each query caption embedding 210 (e.g., more than one sound tag associated with each detected sound) and a tag embedding 514 associated with a media file 152 may be compared to each query tag embedding 410 of a plurality of query tag embeddings 410, and a representative distance being determined, such as an average distance between the tag embedding 514 and each of the plurality of query tag embeddings 410. A sum of similarity metrics corresponding to a tag embedding is determined as part of calculating the average similarity metric. All similarity metrics corresponding to a tag embedding are weighted the same relative to each other.); and
determining the first description objects according to the target similarity of each second semantic vector (see e.g., [0066] for the search results 124 being limited to include only a specific number of most similar results and [0089] for the query 120 being used to generate multiple types of embeddings, which are compared to corresponding embeddings of the file repository 150 and for example, a query caption embedding 210 being compared to the caption embeddings 162 and query tag embedding(s) 410 being compared to tag embeddings 514 associated with the media files 152, when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, and a second weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. The top-ranked captions are determined based on the rankings of both caption embeddings and tag embeddings. The rankings of the tag embeddings are based on the sum of similarity metrics corresponding to each tag embedding.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein there are multiple candidate vector sets, and different candidate vector sets are respectively obtained by performing vectorization processing on the query sentence based on different semantic vector models, wherein determining the first description objects according to the one or more candidate vector sets and the number of the one or more candidate vector sets comprises: performing weighting processing on the initial similarity of each second semantic vector according to a corresponding ranking order of the second semantic vector in each candidate vector set, to obtain a reference similarity corresponding to the second semantic vector; determining a sum of multiple reference similarities corresponding to the second semantic vector as a target similarity of the second semantic vector; and determining the first description objects according to the target similarity of each second semantic vector, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 10, the limitations of parent claims 1, 3, 7, and 8 have been discussed above. Saini does not specifically disclose wherein the preset retrieval library comprises aggregated service data, the aggregated service data comprises multiple service description objects, and the multiple service description objects are associated with the multiple second semantic vectors, wherein determining the first description objects according to the target similarity of each second semantic vector comprises: ranking the multiple second semantic vectors according to the target similarity of each second semantic vector; and taking service description objects corresponding to a second number of second semantic vectors ranked in the front as the first description objects. However, Mahfuz teaches wherein the preset retrieval library comprises aggregated service data [media files] (see e.g., [0081] and FIG. 5 for the file repository 150 including the set of media files 252, and each media file 152 being associated with metadata 154 and sound caption(s) 156A. The device aggregates media files with metadata and captions), the aggregated service data comprises multiple service description objects [captions] (see e.g., [0081] for each media file 152 being associated with sound caption(s) 156A), and the multiple service description objects are associated with the multiple second semantic vectors (see e.g., [0081] for each media file 152 being associated with sound caption(s) 156A and corresponding caption embeddings 162), wherein determining the first description objects according to the target similarity of each second semantic vector comprises:
ranking the multiple second semantic vectors according to the target similarity of each second semantic vector (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156 and the search results 124 identifying one or more media files 152 (or portions of media files 152, such as a portion of a media file 152 associated with a particular sound caption 156) associated with a set of the caption embeddings 160 that are most similar, among the set of media files 252, to the query caption embedding 210. The multiple caption embeddings are ranked according to the weighted similarity metric of each caption embedding.); and
taking service description objects corresponding to a second number of second semantic vectors ranked in the front as the first description objects (see e.g., [0066] for the search results 124 being limited to include only a specific number of most similar results. The captions corresponding to the top-ranked caption embeddings are taken as the top-ranked captions.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the preset retrieval library comprises aggregated service data, the aggregated service data comprises multiple service description objects, and the multiple service description objects are associated with the multiple second semantic vectors, wherein determining the first description objects according to the target similarity of each second semantic vector comprises: ranking the multiple second semantic vectors according to the target similarity of each second semantic vector; and taking service description objects corresponding to a second number of second semantic vectors ranked in the front as the first description objects, as taught by Mahfuz, for the benefit of determining results relevant to a time range (see e.g., Mahfuz, [0057]).
As to claim 11, the limitations of parent claims 1 and 3 have been discussed above. Saini does not specifically disclose wherein determining the candidate description object according to the first description objects and the second description objects comprises: determining a first score value of each first description object and a second score value of each second description object based on a preset scoring model, wherein the first score value is used to describe a matching degree corresponding to the first description object and the query sentence, and the second score value is used to describe a matching degree corresponding to the second description object and the query sentence; ranking the first description objects and the second description objects according to the first score value of each first description object and the second score value of each second description object; and taking a third number of first description objects and/or second description objects ranked in the front as candidate description objects. However, Mahfuz teaches wherein determining the candidate description object according to the first description objects and the second description objects comprises:
determining a first score value [similarity metric for the caption] of each first description object and a second score value [similarity metric for the tag] of each second description object based on a preset scoring model [machine learning model], wherein the first score value is used to describe a matching degree corresponding to the first description object and the query sentence, and the second score value is used to describe a matching degree corresponding to the second description object and the query sentence (see e.g., [0051] for the sound captioning engine 146 including one or more machine-learning models and for example, the sound captioning engine 146 including one or more embedding generators 142, each of which corresponds to or includes at least one trained machine-learning model, [0083] for the natural language processor 302 being operable to determine the target sound description 304 from text of the query 120 and to provide the target sound description 304 to the tag embedding generator 408, the caption embedding generator 242, or both, the caption embedding generator 242 generating the query caption embedding 210 based on the target sound description 304, and in some implementations, the tag embedding generator 408 generating the query tag embedding(s) 410 based on the target sound description 304, [0085] for the comparator 140 determining similarity metrics by comparing the query caption embedding 210 and the caption embeddings 162 in the caption embedding space 220, and [0086] for the comparator 140 determining similarity metrics by comparing the tag embedding(s) 410 and the tag embeddings 514 in the tag embedding space 510. The comparator determines a similarity metric for each caption, which describes a matching degree corresponding to the caption and the query. The comparator determines a similarity metric for each tag, which describes a matching degree corresponding to the tag and the query.);
ranking the first description objects and the second description objects according to the first score value of each first description object and the second score value of each second description object (see e.g., [0089] for when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, a second weight being applied to similarity values based on comparisons of the audio embeddings 530 to the query audio embedding(s) 406, and a third weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. Captions and tags are ranked based on their respective similarity metrics.); and
taking a third number of first description objects and/or second description objects ranked in the front as candidate description objects (see e.g., [0066] for the search results 124 being limited to include only a specific number of most similar results. The top-ranked captions are determined.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein determining the candidate description object according to the first description objects and the second description objects comprises: determining a first score value of each first description object and a second score value of each second description object based on a preset scoring model, wherein the first score value is used to describe a matching degree corresponding to the first description object and the query sentence, and the second score value is used to describe a matching degree corresponding to the second description object and the query sentence; ranking the first description objects and the second description objects according to the first score value of each first description object and the second score value of each second description object; and taking a third number of first description objects and/or second description objects ranked in the front as candidate description objects, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 12, the limitations of parent claims 1, 3, and 11 have been discussed above. Saini teaches
obtaining sample data, wherein, the sample data comprises: a sample query sentence [query] and multiple sample replay sentences [model-generated output results] corresponding to the sample query sentence (see e.g., [0039] for another metadata model identifying one or more passages of text that are semantically and/or lexically related to at least a portion of the input query 104 and [0072] for each training example identifying an example of input information to be fed to a model (such as a query), together with a ground-truth output result that is accepted as a correct transformation of the input information, for example, in the case of a metadata model, a training example including an input query 104 and a metadata item that is considered an appropriate match for the metadata item, during training of each model, the training system 802 producing a model-generated output result for each instance of input information in its set of training examples, and the training system 802 uses a loss function (e.g., a cross entropy loss function) to compute the difference between the model-generated output result and the ground-truth output result. A query is obtained. Model-generated output results corresponding to the query are obtained. The model-generated output results include passages of text.); and
determining a service scenario [ground-truth output] corresponding to the query sentence (seee.g., [0072] for each training example identifying an example of input information to be fed to a model (such as a query), together with a ground-truth output result that is accepted as a correct transformation of the input information and for example, in the case of a metadata model, a training example including an input query 104 and a metadata item that is considered an appropriate match for the metadata item. A ground-truth output corresponding to the query is determined.).
Saini does not specifically disclose wherein the preset scoring model is obtained by training in a following manner: labeling the multiple sample replay sentences according to the service scenario to determine positive sample reply sentences and negative sample reply sentences from the multiple sample reply sentences; and training an initial scoring model based on the sample query sentence, the positive sample reply sentences and the negative sample reply sentences to obtain the preset scoring model. However, Mahfuz teaches wherein the preset scoring model is obtained by training in a following manner:
labeling the multiple sample replay sentences [predicted captions] according to the service scenario [ground truth captions] to determine positive sample reply sentences [predicted caption sentences in entailment of ground truth caption sentences] and negative sample reply sentences [predicted caption sentences in contradiction to ground truth caption sentences] from the multiple sample reply sentences (see e.g., [0096] for caption embedding generator 688 generating a predicted caption embedding 630, [0099] for the difference calculation 634 being based on a comparison of a predicted caption embedding 630 and a ground truth caption embedding 632 for a particular sound and the ground truth caption embedding 632 for each sound being determined by the caption embedding generator 688, and [0105] for Sentence-BERT being used as the caption embedding generator 688 and being configured or trained to distinguish whether two sentences are in entailment of, in contradiction to, or neutral with respect to each other. Predicted captions are labelled according to ground truth captions as being in entailment of the ground truth captions or in contradiction to the ground truth captions.); and
training an initial scoring model [machine learning model] based on the sample query sentence, the positive sample reply sentences and the negative sample reply sentences to obtain the preset scoring model (see e.g., [0091] for during the training process illustrated in FIG. 6, a loss calculator 640 determining a loss metric based on one or more of a plurality of difference calculations (“Diff. Calc.” in FIG. 6), the training process being iterative, and during each iteration, changes in the loss metric being used to adjust machine-learning parameters of one or more of the machine-learning models of the sound captioning engine 670, and for example, a machine-learning optimizer 650 using one or more backpropagation operations (or another machine-learning optimization process) to adjust the machine-learning parameters of the machine-learning model(s) to reduce the loss metric. A machine learning model is trained based on the query, the predicted caption sentences in entailment of ground truth caption sentences, and the predicted caption sentences in contradiction to ground truth caption sentences to obtain the trained machine learning model.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the preset scoring model is obtained by training in a following manner: labeling the multiple sample replay sentences according to the service scenario to determine positive sample reply sentences and negative sample reply sentences from the multiple sample reply sentences; and training an initial scoring model based on the sample query sentence, the positive sample reply sentences and the negative sample reply sentences to obtain the preset scoring model, as taught by Mahfuz, for the benefit of employing a generator specifically enabled to distinguish sentences (see e.g., Mahfuz, [0105]).
As to claim 14, Saini teaches an electronic device, comprising:
at least on processor (see e.g., [0085] for the computing system 1402 including a processing system 1404 including one or more processors); and
a memory communicatively coupled to the at least one processor (see e.g., [0088] for the computing system 1402 utilizing any instance of the computer-readable storage media 1406 in different ways and for example, in some implementations, any instance of the computer-readable storage media 1406 representing a hardware memory unit (such as random access memory (RAM)) for storing information during execution of a program by the computing system 1402); wherein,
the memory is stored with instructions executable by the at least one processor (see e.g., [0088] for computer-readable storage media 1406 representing a hardware memory unit (such as random access memory (RAM)) for storing information during execution of a program by the computing system 1402), wherein the at least one processor is configured to:
obtain a query sentence [input query], wherein the query sentence has at least one candidate description object [query embedding] (see e.g., [0079] for the computing system 102 receiving the input query 104 and the computing system 102 encoding the input query 104 into a query embedding EQ (e.g., the query embedding 228). The computing system obtains an input query, which has a query embedding.);
perform semantic recognition on the query sentence to obtain first semantic information [metadata embedding representing user location] corresponding to the query sentence (see e.g., [0028] for a first metadata model 202 (M1) transforming the input query 104 to a first lookup vector 204, and then using the first lookup vector 204 to identify a first metadata embedding EM1 in a first metadata vector space 206 and each of the above-identified metadata embeddings being associated with a particular metadata item, [0030] for embeddings in each vector space having a relationship to each other and in particular, the distance between any pair of vectors in a single vector space reflecting the semantic similarity between the concepts associated with the two vectors, [0036] for different applications using metadata models that produce different kinds of metadata, and [0041] for metadata model receiving location information that identifies the location associated with the user and the metadata model leveraging location information in various ways, such as by adding text to the input query 104 that identifies the location of the user, and then transforming the thus-augmented query into a query embedding. The computing system obtains a metadata embedding representing location of the querying user by recognizing that the embedding is semantically similar to the input query augmented with location information.);
input the at least one candidate description object and the first semantic information into a large model [generative model], to identify a target description object [keyword] from the at least one candidate description object based on the large model (see e.g., [0024] for a query-augmenting component 110 combining the query embedding with the metadata embedding(s), to produce an augmented embedding EQ+ and a retrieval component 112 using the augmented embedding to identify one or more target items and [0049] for a retrieval component 602 that represents the retrieval component 112 of FIG. 1, the retrieval component 602 using a machine-trained generative model 604 to generate one or more target items based on the augmented embedding EQ+, for example, the machine-trained generative model 604 treating the augmented embedding as a prompt, and using a language model of any kind to autoregressively transform the prompt into at least one item, token by token, and for instance, the retrieval component 602 transforming the augmented embedding into a keyword (which includes one or more words), and then finds an ad associated with the generated keyword. The query embedding and the metadata embedding are input into a generative model to identify a keyword from the input query based on the generative model.);
select target service data [ad] from at least one piece of service data [target data] corresponding to the target description object (see e.g., [0049] for a retrieval component 602 that represents the retrieval component 112 of FIG. 1, the retrieval component 602 using a machine-trained generative model 604 to generate one or more target items based on the augmented embedding EQ+, for example, the machine-trained generative model 604 treating the augmented embedding as a prompt, and using a language model of any kind to autoregressively transform the prompt into at least one item, token by token, and for instance, the retrieval component 602 transforming the augmented embedding into a keyword (which includes one or more words), and then finding an ad associated with the generated keyword. An ad is selected from the target data corresponding to the keyword.); and
generate a reply sentence [output information] corresponding to the query sentence according to the target service data (see e.g., [0024] for an output-generating component 116 providing output information 118 that includes each identified target item and the output-generating component 116 producing output information 118 which contains an ad that matches the input query 104. Output information corresponding to the input query is generated according to the ad. Output information may encompass a sentence within the ad in reply to the input query.).
Saini does not specifically disclose perform semantic recognition on the query sentence to obtain second semantic information corresponding to the query sentence, wherein categories of the first semantic information and the second semantic information are different; and select target service data from at least one piece of service data corresponding to the target description object based on the second semantic information. However, Mahfuz teaches
perform semantic recognition on the query sentence [query] to obtain second semantic information [context term indicating a range of time stamps] corresponding to the query sentence, wherein categories of the first semantic information and the second semantic information are different (see e.g., [0057] for the query 120 including “where is that video of the bell ringing multiple times that I took last week?” and “last week” being a context term indicating a range of time stamps when the target media file was created and [0069] for the natural language processor 302 being configured to process the text query 202 to identify the portion of the text query 202 that corresponds to the target sound description 304 and the portion of the text query 202 that corresponds to the context term(s) 306. The natural language processor performs semantic recognition on the query to obtain a context term indicating a range of time stamps, which corresponds to the query. A range of time stamps is a different category from location of the querying user.); and
select target service data [highest-ranked search results] from at least one piece of service data [media file] corresponding to the target description object [sound caption] based on the second semantic information (see e.g., [0034] for if the query includes natural-language text, at least a portion of the text being used to generate a query caption embedding (e.g., a sentence embedding) and in this example, the query caption embedding being compared, in a caption embedding space, to caption embeddings associated with the media files to determine a similarity metric, and the search results being determined based on the similarity metric, [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156, [0066] for the search results 124 being limited to include only a specific number of most similar results, [0070] for filter criteria being based on the context term(s) 306 and filter criteria including a time criterion, and [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124 and for example, in some such implementations, the comparator 140 determining similarity metrics for caption embeddings 160 associated with metadata 154 that failed to satisfy the filter criteria; however, such caption embeddings being unfavorably weighted to appear lower in the ranked search results 124 than they would had their metadata 154 satisfied the filter criteria. The highest-ranked search results are selected from at least one media file corresponding to the sound caption based on the context term indicating a range of time stamps).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini to perform semantic recognition on the query sentence to obtain second semantic information corresponding to the query sentence, wherein categories of the first semantic information and the second semantic information are different; and select target service data from at least one piece of service data corresponding to the target description object based on the second semantic information, as taught by Mahfuz, for the benefit of determining results relevant to a time range (see e.g., Mahfuz, [0057]).
As to claim 17, the limitations of parent claim 14 have been discussed above. Saini teaches
wherein the first semantic information is semantic information of a geographic area dimension (see e.g., [0041] for metadata model receiving location information that identifies the location associated with the user and the metadata model leveraging location information in various ways, such as by adding text to the input query 104 that identifies the location of the user, and then transforming the thus-augmented query into a query embedding. Location of a querying user is a geographic area dimension.).
Saini does not specifically disclose wherein the second semantic information is semantic information of a statistical frequency dimension related to the candidate description object, and/or semantic information of a statistical time dimension related to the candidate description object. However, Mahfuz teaches
wherein the second semantic information is semantic information of a statistical frequency dimension related to the candidate description object, and/or semantic information of a statistical time dimension related to the candidate description object [caption] (see e.g., [0057] for the query 120 including “where is that video of the bell ringing multiple times that I took last week?” and “last week” being a context term indicating a range of time stamps when the target media file was created, [0069] for the natural language processor 302 being configured to process the text query 202 to identify the portion of the text query 202 that corresponds to the target sound description 304 and the portion of the text query 202 that corresponds to the context term(s) 306, and [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124 and for example, in some such implementations, the comparator 140 determining similarity metrics for caption embeddings 160 associated with metadata 154 that failed to satisfy the filter criteria; however, such caption embeddings being unfavorably weighted to appear lower in the ranked search results 124 than they would had their metadata 154 satisfied the filter criteria. A range of timestamps is a statistical time dimension related to the caption corresponding to the caption embedding).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the second semantic information is semantic information of a statistical frequency dimension related to the candidate description object, and/or semantic information of a statistical time dimension related to the candidate description object, as taught by Mahfuz, for the benefit of determining results relevant to a time range (see e.g., Mahfuz, [0057]).
As to claim 18, the limitations of parent claim 14 have been discussed above. Saini does not specifically disclose wherein the at least one candidate description object is determined based on a following manner: obtaining one or more first semantic vectors and a first keyword corresponding to the query sentence; determining first description objects matching the one or more first semantic vectors and second description objects matching the first keyword from a preset retrieval library; and determining the candidate description object according to the first description objects and the second description objects. However, Mahfuz teaches wherein the at least one candidate description object is determined based on a following manner:
obtaining one or more first semantic vectors [query caption embedding] and a first keyword [word from target sound description] corresponding to the query sentence (see e.g., [0083] for the natural language processor 302 being operable to determine the target sound description 304 from text of the query 120 and to provide the target sound description 304 to the tag embedding generator 408, the caption embedding generator 242, or both, in this example, the caption embedding generator 242 generating the query caption embedding 210 based on the target sound description 304, and in some implementations, the tag embedding generator 408 generating the query tag embedding(s) 410 based on the target sound description 304, and the caption embedding generator 242 generating the query caption embedding 210 based on the query tag embedding(s). A query caption embedding and a word from the target sound description corresponding to the query are obtained.);
determining first description objects [captions] matching the one or more first semantic vectors and second description objects [sound tags] matching the first keyword from a preset retrieval library [file repository] (see e.g., [0083] for the comparator 140 determining similarity metrics by comparing the query caption embedding 210 and the caption embeddings 162, [0084] for in implementations in which the target sound description 304 is processed by the tag embedding generator 408 to generate the query tag embedding(s) 410, the comparator 140 also determining similarity metrics by comparing the query tag embedding(s) 410 and the tag embeddings 514 in a tag embedding space, [0089] for search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and FIG. 5 for the file repository including captions and sound tags. Comparison is used to determine captions and sound tags corresponding to search results. The captions match the query caption embedding and the sound tags match the target sound description.); and
determining the candidate description object [highly-ranked caption or sound tag] according to the first description objects and the second description objects (see e.g., [0089] for when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, a second weight being applied to similarity values based on comparisons of the audio embeddings 530 to the query audio embedding(s) 406, and a third weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. A highly ranked caption or sound tag is determined according to the caption and sound tags.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the at least one candidate description object is determined based on a following manner: obtaining one or more first semantic vectors and a first keyword corresponding to the query sentence; determining first description objects matching the one or more first semantic vectors and second description objects matching the first keyword from a preset retrieval library; and determining the candidate description object according to the first description objects and the second description objects, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 19, the limitations of parent claims 14 and 18 have been discussed above. Saini does not specifically disclose wherein the preset retrieval library is constructed based on a following manner: obtaining a service database, wherein the service database comprises multiple pieces of initial service data; determining service description objects of the initial service data, and data record types and data record frequencies corresponding to the service description objects; performing aggregation processing on the multiple pieces of initial service data based on the service description objects, the data record types, and the data recording frequencies, to obtain aggregated service data, wherein the aggregated service data comprises the at least one piece of service data corresponding to the target description object; performing vectorization processing on the service description objects based on a semantic vector model, to obtain a second semantic vector; and constructing the preset retrieval library based on the aggregated service data and the second semantic vector. However, Mahfuz teaches wherein the preset retrieval library is constructed based on a following manner:
obtaining a service database [set of media files], wherein the service database comprises multiple pieces of initial service data [media files] (see e.g., [0044] for the system 100 including a device 102 that includes one or more processors 190 and memory 192 and the memory 192 including a file repository 150 storing media files 152, [0080] for the media search engine 130 searching the file repository 150 based on the query 120, and [0081] for the file repository 150 including the set of media files 252. The device obtains a set of media files, which includes multiple media files.);
determining service description objects [captions] of the initial service data, and data record types [file types] and data record frequencies [sound frequency] corresponding to the service description objects (see e.g., [0057] for the query 120 including “where is that video of the bell ringing multiple times that I took last week?” and the term “video” being a context term indicating a file type of a target media file, [0070] for filter criteria being based on the context term(s) 306 and filter criteria including a target file type, [0073] for the filter criteria being used to determine weights applied to the similarity metrics used to sort the search results 124, and [0075] for the sound captioning engine 146 being configured to generate sound captions descriptive of sounds in the media files 152. Captions, which may include the fact that a sound occurred multiple times, are determined by the sound captioning engine. A file type corresponding to the caption is determined for weighting.);
performing aggregation processing on the multiple pieces of initial service data based on the service description objects, the data record types, and the data recording frequencies, to obtain aggregated service data, wherein the aggregated service data comprises the at least one piece of service data corresponding to the target description object (see e.g., [0081] and FIG. 5 for the file repository 150 including the set of media files 252, and each media file 152 being associated with metadata 154 and sound caption(s) 156A. The device aggregates media files with metadata, which includes file type, and captions, which includes sound frequencies. The aggregated media files include at least one media file corresponding to the target sound caption.);
performing vectorization processing [embedding] on the service description objects based on a semantic vector model [machine-learning model corresponding to caption embedding generator], to obtain a second semantic vector [caption embedding] (see e.g., [0051] for the sound captioning engine 146 including one or more machine-learning models and for example, the sound captioning engine 146 including one or more embedding generators 142, each of which corresponds to or includes at least one trained machine-learning model, [0052] for the sound captioning engine 146 being operable to generate embeddings 158 representing various data stored in the file repository 150 and for example, when a new media file 152 is added to the file repository 150, the sound captioning engine 146 generating a caption embedding 160 representing a sound caption 156 of the new media file 152, [0054] for each of the caption embeddings 160 being viewed as a vector designating a particular location in a caption embedding space. Embedding is performed on captions based on a machine-learning model corresponding to caption embedding generator to obtain a caption embedding, which is a vector.); and
constructing the preset retrieval library based on the aggregated service data and the second semantic vector (see e.g., [0081] and FIG. 5 for the file repository 150 including the set of media files 252, and each media file 152 being associated with metadata 154, sound caption(s) 156A, and corresponding caption embeddings 162.The file repository is constructed from media files with metadata, which includes file type, captions, which includes sound frequencies, and caption embeddings.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the preset retrieval library is constructed based on a following manner: obtaining a service database, wherein the service database comprises multiple pieces of initial service data; determining service description objects of the initial service data, and data record types and data record frequencies corresponding to the service description objects; performing aggregation processing on the multiple pieces of initial service data based on the service description objects, the data record types, and the data recording frequencies, to obtain aggregated service data, wherein the aggregated service data comprises the at least one piece of service data corresponding to the target description object; performing vectorization processing on the service description objects based on a semantic vector model, to obtain a second semantic vector; and constructing the preset retrieval library based on the aggregated service data and the second semantic vector, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 22, the limitations of parent claims 14 and 18 have been discussed above. Saini does not specifically disclose wherein the preset retrieval library comprises: multiple second semantic vectors, wherein the at least one processor is configured to: determine an initial similarity between a first semantic vector and each second semantic vector in the preset retrieval library; rank the multiple second semantic vectors according to initial similarities; construct a candidate vector set corresponding to the first semantic vector according to a first number of second semantic vectors ranked in the front; and determine the first description objects according to one or more candidate vector sets and a number of the one or more candidate vector sets. However, Mahfuz teaches wherein the preset retrieval library comprises:
multiple second semantic vectors (see e.g., [0081] and FIG. 5 for the file repository 150 including caption embeddings 162), wherein the at least one processor is configured to:
determine an initial similarity [similarity metric] between a first semantic vector and each second semantic vector in the preset retrieval library (see e.g., [0064] for the comparator 140 being configured to determine the similarity metric for each caption embedding 160 based on a distance 214, in the caption embedding space 220, between the caption embedding 160 and the query caption embedding 210 and the distance 214 being determined, for example, as a cosine distance, a Euclidean distance, or based on some other distance measure. A similarity metric is determined between a query caption embedding and each caption embedding in the file repository.);
rank the multiple second semantic vectors according to initial similarities (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156, the search results 124 including a rank sorted list of results, and in such implementations, the search results 124 being sorted based on their respective similarity metrics. The multiple caption embeddings are ranked according to similarity metrics.);
construct a candidate vector set [top-ranked caption embeddings] corresponding to the first semantic vector according to a first number of second semantic vectors ranked in the front (see e.g., [0066] for the search results 124 being limited to include only a specific number of most similar results. The top-ranked caption embeddings corresponding to the query caption embedding are determined.); and
determine the first description objects according to one or more candidate vector sets [top-ranked caption embeddings or top-ranked caption embeddings and top-ranked tag embeddings] and a number of the one or more candidate vector sets (see e.g., [0065] for the media search engine 130 generating the search results 124 based on similarity metrics associated with the sound captions 156 and the search results 124 identifying one or more media files 152 (or portions of media files 152, such as a portion of a media file 152 associated with a particular sound caption 156) associated with a set of the caption embeddings 160 that are most similar, among the set of media files 252, to the query caption embedding 210 and [0089] for the query 120 being used to generate multiple types of embeddings, which are compared to corresponding embeddings of the file repository 150 and for example, a query caption embedding 210 being compared to the caption embeddings 162 and query tag embedding(s) 410 being compared to tag embeddings 514 associated with the media files 152, when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, and a second weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. Top-ranked captions are determined based on top-ranked caption embeddings. Top-ranked captions are also determined based on whether only caption embeddings are being considered (one set of embeddings) or whether both caption embeddings and tag embeddings are being considered (two sets of embeddings).).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the preset retrieval library comprises: multiple second semantic vectors, wherein the at least one processor is configured to: determine an initial similarity between a first semantic vector and each second semantic vector in the preset retrieval library; rank the multiple second semantic vectors according to initial similarities; construct a candidate vector set corresponding to the first semantic vector according to a first number of second semantic vectors ranked in the front; and determine the first description objects according to one or more candidate vector sets and a number of the one or more candidate vector sets, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
As to claim 23, the limitations of parent claims 14 and 18 have been discussed above. Saini does not specifically disclose wherein the at least on processor is configured to: determine a first score value of each first description object and a second score value of each second description object based on a preset scoring model, wherein the first score value is used to describe a matching degree corresponding to the first description object and the query sentence, and the second score value is used to describe a matching degree corresponding to the second description object and the query sentence; rank the first description objects and the second description objects according to the first score value of each first description object and the second score value of each second description object; and take a third number of first description objects and/or second description objects ranked in the front as candidate description objects. However, Mahfuz teaches wherein the at least one processor is configured to:
determine a first score value [similarity metric for the caption] of each first description object and a second score value [similarity metric for the tag] of each second description object based on a preset scoring model [machine learning model], wherein the first score value is used to describe a matching degree corresponding to the first description object and the query sentence, and the second score value is used to describe a matching degree corresponding to the second description object and the query sentence (see e.g., [0051] for the sound captioning engine 146 including one or more machine-learning models and for example, the sound captioning engine 146 including one or more embedding generators 142, each of which corresponds to or includes at least one trained machine-learning model, [0083] for the natural language processor 302 being operable to determine the target sound description 304 from text of the query 120 and to provide the target sound description 304 to the tag embedding generator 408, the caption embedding generator 242, or both, the caption embedding generator 242 generating the query caption embedding 210 based on the target sound description 304, and in some implementations, the tag embedding generator 408 generating the query tag embedding(s) 410 based on the target sound description 304, [0085] for the comparator 140 determining similarity metrics by comparing the query caption embedding 210 and the caption embeddings 162 in the caption embedding space 220, and [0086] for the comparator 140 determining similarity metrics by comparing the tag embedding(s) 410 and the tag embeddings 514 in the tag embedding space 510. The comparator determines a similarity metric for each caption, which describes a matching degree corresponding to the caption and the query. The comparator determines a similarity metric for each tag, which describes a matching degree corresponding to the tag and the query.);
rank the first description objects and the second description objects according to the first score value of each first description object and the second score value of each second description object (see e.g., [0089] for when multiple types of embeddings are compared to corresponding embeddings of the file repository 150, each comparison generating one or more search results, in such implementations, search results based on comparisons of different types of embeddings being weighted differently to generate a ranked listing of the search results 124, and to illustrate, to sort the search results 124 based on a ranking, a first weight being applied to similarity values based on comparisons of the caption embeddings 162 to the query caption embedding 210, a second weight being applied to similarity values based on comparisons of the audio embeddings 530 to the query audio embedding(s) 406, and a third weight being applied to similarity values based on comparisons of the tag embeddings 514 to the query tag embedding(s) 410. Captions and tags are ranked based on their respective similarity metrics.); and
take a third number of first description objects and/or second description objects ranked in the front as candidate description objects (see e.g., [0066] for the search results 124 being limited to include only a specific number of most similar results. The top-ranked captions are determined.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini wherein the at least one processor is configured to: determine a first score value of each first description object and a second score value of each second description object based on a preset scoring model, wherein the first score value is used to describe a matching degree corresponding to the first description object and the query sentence, and the second score value is used to describe a matching degree corresponding to the second description object and the query sentence; rank the first description objects and the second description objects according to the first score value of each first description object and the second score value of each second description object; and take a third number of first description objects and/or second description objects ranked in the front as candidate description objects, as taught by Mahfuz, for the benefit of representing search content with multiple types of embeddings (see e.g., Mahfuz, [0089]).
Claims 5, 6, 20, and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Saini et al. (US Publication No. 2025/0061136) in view of Mahfuz et al. (US Publication No. 2024/0134908) as applied to claims 1-4, 7-12, 14, 17-19, 22, and 23 above, and further in view of Sundaram et al. (US Publication No. 2025/0077769).
As to claim 5, the limitations of parent claims 1, 3, and 4 have been discussed above. Saini in view of Mahfuz does not specifically disclose wherein performing the vectorization processing on the service description objects based on the semantic vector model, to obtain the second semantic vector comprises: performing the vectorization processing on the service description objects based on a monolingual semantic vector model, to obtain the second semantic vector; and/or performing the vectorization processing on the service description objects based on a multilingual semantic vector model, to obtain the second semantic vector. However, Sundaram teaches wherein performing the vectorization processing on the service description objects based on the semantic vector model, to obtain the second semantic vector comprises:
performing the vectorization processing on the service description objects based on a monolingual semantic vector model, to obtain the second semantic vector; and/or
performing the vectorization processing [encoding] on the service description objects [tokens] based on a multilingual semantic vector model [multilingual encoder model], to obtain the second semantic vector [contextual embedding output] (see e.g., [0048] for an example architecture of a multilingual encoder model being depicted in FIG. 6, which leverages a multilingual vocabulary and multilingual embeddings to capture the meaning of words in multiple languages, as shown, the example input text being the German text, “Im osten geht die sonne auf,” the input text being divided into individual tokens (e.g., “Im”, “osten”, “geht”, “die”, “sonne”, “auf”), and the corresponding token indexes being looked up in the multilingual vocabulary, the token indexes being then passed to the embedding layer, which generates multilingual embedding values (e.g., in the form of floating point values) that serve as inputs to the encoder system to produce a contextual embeddings output, the embedding layer transforming the token indexes associated with the input text into embeddings (e.g., dense vector representation), which may capture semantic and contextual information of words in a continuous vector space, for example, the embedding layer being embodied as a matrix where each row corresponds with a token index in the vocabulary, and the token index itself being associated with an n-dimensional vector, the encoder system including one or more layers of a neural network or other machine learning technology, and transforming the sequence of input embedding vectors into a fixed length vector representation that captures the overall information of the sequence, and the contextual vector generated by the encoder system being independent of any specific language, as the multilingual embedding output may capture semantic and contextual information across different languages. Encoding is performed on the tokens based on a multilingual encoder model to obtain contextual embedding output, which is a vector.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini in view of Mahfuz wherein performing the vectorization processing on the service description objects based on the semantic vector model, to obtain the second semantic vector comprises: performing the vectorization processing on the service description objects based on a monolingual semantic vector model, to obtain the second semantic vector; and/or performing the vectorization processing on the service description objects based on a multilingual semantic vector model, to obtain the second semantic vector, as taught by Sundaram, for the benefit of capturing the meaning of words in multiple languages (see e.g., Sundaram, [0048]).
As to claim 6, the limitations of parent claims 1 and 3 have been discussed above. Saini in view of Mahfuz does not specifically disclose wherein obtaining the one or more first semantic vectors corresponding to the query sentence comprises: performing vectorization processing on the query sentence based on a monolingual semantic vector model, to obtain a first semantic vector; and/or performing the vectorization processing on the query sentence based on a multilingual semantic vector model, to obtain a first semantic vector. However, Sundaram teaches wherein obtaining the one or more first semantic vectors corresponding to the query sentence comprises:
performing vectorization processing on the query sentence based on a monolingual semantic vector model, to obtain a first semantic vector; and/or
performing the vectorization processing [encoding] on the query sentence [tokens] based on a multilingual semantic vector model [multilingual encoder model], to obtain a first semantic vector [contextual embedding output] (see e.g., [0048] for an example architecture of a multilingual encoder model being depicted in FIG. 6, which leverages a multilingual vocabulary and multilingual embeddings to capture the meaning of words in multiple languages, as shown, the example input text being the German text, “Im osten geht die sonne auf,” the input text being divided into individual tokens (e.g., “Im”, “osten”, “geht”, “die”, “sonne”, “auf”), and the corresponding token indexes being looked up in the multilingual vocabulary, the token indexes being then passed to the embedding layer, which generates multilingual embedding values (e.g., in the form of floating point values) that serve as inputs to the encoder system to produce a contextual embeddings output, the embedding layer transforming the token indexes associated with the input text into embeddings (e.g., dense vector representation), which may capture semantic and contextual information of words in a continuous vector space, for example, the embedding layer being embodied as a matrix where each row corresponds with a token index in the vocabulary, and the token index itself being associated with an n-dimensional vector, the encoder system including one or more layers of a neural network or other machine learning technology, and transforming the sequence of input embedding vectors into a fixed length vector representation that captures the overall information of the sequence, and the contextual vector generated by the encoder system being independent of any specific language, as the multilingual embedding output may capture semantic and contextual information across different languages. Encoding is performed on the tokens based on a multilingual encoder model to obtain contextual embedding output, which is a vector.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini in view of Mahfuz wherein obtaining the one or more first semantic vectors corresponding to the query sentence comprises: performing vectorization processing on the query sentence based on a monolingual semantic vector model, to obtain a first semantic vector; and/or performing the vectorization processing on the query sentence based on a multilingual semantic vector model, to obtain a first semantic vector, as taught by Sundaram, for the benefit of capturing the meaning of words in multiple languages (see e.g., Sundaram, [0048]).
As to claim 20, the limitations of parent claims 14, 18, and 19 have been discussed above. Saini in view of Mahfuz does not specifically disclose wherein the at least one processor is configured to: perform the vectorization processing on the service description objects based on a monolingual semantic vector model, to obtain the second semantic vector; and/or perform the vectorization processing on the service description objects based on a multilingual semantic vector model, to obtain the second semantic vector. However, Sundaram teaches wherein the at least one processor is configured to:
perform the vectorization processing on the service description objects based on a monolingual semantic vector model, to obtain the second semantic vector; and/or
perform the vectorization processing [encoding] on the service description objects [tokens] based on a multilingual semantic vector model [multilingual encoder model], to obtain the second semantic vector [contextual embedding output] (see e.g., [0048] for an example architecture of a multilingual encoder model being depicted in FIG. 6, which leverages a multilingual vocabulary and multilingual embeddings to capture the meaning of words in multiple languages, as shown, the example input text being the German text, “Im osten geht die sonne auf,” the input text being divided into individual tokens (e.g., “Im”, “osten”, “geht”, “die”, “sonne”, “auf”), and the corresponding token indexes being looked up in the multilingual vocabulary, the token indexes being then passed to the embedding layer, which generates multilingual embedding values (e.g., in the form of floating point values) that serve as inputs to the encoder system to produce a contextual embeddings output, the embedding layer transforming the token indexes associated with the input text into embeddings (e.g., dense vector representation), which may capture semantic and contextual information of words in a continuous vector space, for example, the embedding layer being embodied as a matrix where each row corresponds with a token index in the vocabulary, and the token index itself being associated with an n-dimensional vector, the encoder system including one or more layers of a neural network or other machine learning technology, and transforming the sequence of input embedding vectors into a fixed length vector representation that captures the overall information of the sequence, and the contextual vector generated by the encoder system being independent of any specific language, as the multilingual embedding output may capture semantic and contextual information across different languages. Encoding is performed on the tokens based on a multilingual encoder model to obtain contextual embedding output, which is a vector.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini in view of Mahfuz wherein the at least one processor is configured to: perform the vectorization processing on the service description objects based on a monolingual semantic vector model, to obtain the second semantic vector; and/or perform the vectorization processing on the service description objects based on a multilingual semantic vector model, to obtain the second semantic vector, as taught by Sundaram, for the benefit of capturing the meaning of words in multiple languages (see e.g., Sundaram, [0048]).
As to claim 21, the limitations of parent claims 14 and 18 have been discussed above. Saini in view of Mahfuz does not specifically disclose wherein the at least one processor is configured to: perform vectorization processing on the query sentence based on a monolingual semantic vector model, to obtain a first semantic vector; and/or perform the vectorization processing on the query sentence based on a multilingual semantic vector model, to obtain a first semantic vector. However, Sundaram teaches wherein the at least one processor is configured to:
perform vectorization processing on the query sentence based on a monolingual semantic vector model, to obtain a first semantic vector; and/or
perform the vectorization processing [encoding] on the query sentence [tokens] based on a multilingual semantic vector model [multilingual encoder model], to obtain a first semantic vector [contextual embedding output] (see e.g., [0048] for an example architecture of a multilingual encoder model being depicted in FIG. 6, which leverages a multilingual vocabulary and multilingual embeddings to capture the meaning of words in multiple languages, as shown, the example input text being the German text, “Im osten geht die sonne auf,” the input text being divided into individual tokens (e.g., “Im”, “osten”, “geht”, “die”, “sonne”, “auf”), and the corresponding token indexes being looked up in the multilingual vocabulary, the token indexes being then passed to the embedding layer, which generates multilingual embedding values (e.g., in the form of floating point values) that serve as inputs to the encoder system to produce a contextual embeddings output, the embedding layer transforming the token indexes associated with the input text into embeddings (e.g., dense vector representation), which may capture semantic and contextual information of words in a continuous vector space, for example, the embedding layer being embodied as a matrix where each row corresponds with a token index in the vocabulary, and the token index itself being associated with an n-dimensional vector, the encoder system including one or more layers of a neural network or other machine learning technology, and transforming the sequence of input embedding vectors into a fixed length vector representation that captures the overall information of the sequence, and the contextual vector generated by the encoder system being independent of any specific language, as the multilingual embedding output may capture semantic and contextual information across different languages. Encoding is performed on the tokens based on a multilingual encoder model to obtain contextual embedding output, which is a vector.).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to modify the computing system of Saini in view of Mahfuz wherein the at least one processor is configured to: perform vectorization processing on the query sentence based on a monolingual semantic vector model, to obtain a first semantic vector; and/or perform the vectorization processing on the query sentence based on a multilingual semantic vector model, to obtain a first semantic vector, as taught by Sundaram, for the benefit of capturing the meaning of words in multiple languages (see e.g., Sundaram, [0048]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Pendar et al. (US Patent No. 12,248,467) for “In some examples, the evaluation prompt can further include instructions to perform statistical calculations based on one or more attributes of the records. For example, the evaluation prompt can be “Rank the data below using the text as context. Return as a table, add ranking as a new column. If price (or other quantitative data) is available for the records, provide median value of the price (or other quantitative data).” Responsive to this evaluation prompt, besides generating the new table 730, the LLM can further output “Median price: $150.00” for the above example” (see col. 15, lines 21-30).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DARA J GLASSER whose telephone number is (571)270-3666. The examiner can normally be reached Monday-Thursday, 10:00am-2:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Apu Mofiz can be reached at (571)272-4080. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
04-14-2026
/DARA J GLASSER/Examiner, Art Unit 2161
/APU M MOFIZ/Supervisory Patent Examiner, Art Unit 2161
