DETAILED ACTION
Claims 1-20 are presented for examination
This office action is in response to submission of application on 17-SEPTEMBER-2022.
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 .
Response to Amendment
The amendment filed on 22-JULY-2026 in response to the non-final office action mailed 15-MAY-2026 has been entered. Claims 1, 3-10, 12-20 remain pending in the application.
With regards to the 103 rejections, the applicant’s amendments to the claims have not overcome the rejections to claims 1-20 as the former prior art sufficiently teaches the newly added limitations of the amended claims.
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.
Claims 1, 3-5, 10, and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1) in view of LIU (C.N. Pub. No. CN 114548273 A) in view of SONG (U.S. Pub. No. US 20170060855 A1) in view of ZHOU (U.S. Pub. No. US 20220138209 A1)
Regarding claim 1, LEGRAND substantially teaches the claim including:
A method for predicting a next user selection in an electronic user interface, the method comprising: receiving, from a user, a sequence of selections of documents; ([0261] The user model can be designed and implemented to determine a probability that a document D would be chosen, given a set of documents presented to the user and the sequence of selections/clicks up to that point. (the sequence of clicks would be a sequence of selections the users made on documents)) generating, for each document in the sequence, a respective attribute vector, ([0048] In an implementation of the present disclosure, a system can have several aspects, and different implementation need not implement all of the following aspects: 1) a module for creating an initial query, 2) a module for obtaining a set of candidate results satisfying the initial query, 3) a module for determining the distance or similarity between candidate results or a module for embedding the candidate results in a vector space, [0075] In order to embed a document catalog in a vector space each document must be associated with a vector. A distance between two documents in such a space is then determined using standard measures of distance using vectors. [0081] Often embeddings will consider documents in multiple ways. For example, a product may be embedded in terms of the meta-data associated with that product, the image of that product, and the textual content of reviews for that product. Such an embedding may be achieved by developing Kernels for each aspect of the document and combining those Kernels in some way, e.g., via a linear combination. (here, we see that the vector space, and the vectors embedded within, are embedded off of the metadata of the document selected))
While LEGRAND teaches obtaining document selections and metadata vectors, it does not explicitly teach:
Based on a plurality of vector portions, the plurality of vector portions comprising: a numerical attribute vector portion representative of numerical attributes of the document; a category attribute vector portion representative of category information of the document; a text content vector portion representative of text content of the document; and an image content vector portion representative of one or more images in the document; inputting the attribute vectors of the sequence into a machine learning model;
however, in analogous art that similarly makes vectors off of attribute data, LIU teaches:
Based on a plurality of vector portions, the plurality of vector portions comprising : a numerical attribute vector portion representative of numerical attributes of the document; a category attribute vector portion representative of category information of the document; a text content vector portion representative of text content of the document; and an image content vector portion representative of one or more images in the document; ((LIU page 5, paragraph 11) It is understandable that , the category of attribute information of each attribute may include, but is not limited to, text, numbers, images, and so on. (LIU page 5, paragraph 12)When converting the target attribute information into the corresponding original vector, it is possible to determine the target category to which the attribute information corresponding to each attribute in the target attribute information belongs, and use the information conversion algorithm corresponding to the target category to convert the attribute information corresponding to the attribute to the attribute The corresponding representation vector is further spliced with the representation vector corresponding to each attribute, thereby obtaining the original vector corresponding to the data sample. For example, if the target attribute information of the data sample includes attribute information corresponding to attribute 1 and attribute information corresponding to attribute 2, the attribute information corresponding to attribute 1 is converted into the corresponding representation vector 1, and the attribute information corresponding to attribute 2 is converted into corresponding The representation vector 2 is further spliced with the representation vector 1 and the representation vector 2 to obtain the original vector corresponding to the data sample (here it can be seen that the vectors are split into portions of the original vector, thus vector portions.) (LIU page 4 paragraph 19-24) The second determining unit is specifically configured to: respectively obtain the target category to which the attribute information corresponding to each attribute in the target attribute information belongs; Using an information conversion algorithm corresponding to the target category, the attribute information corresponding to the attribute is converted into a representation vector corresponding to the attribute; The representation vectors corresponding to each attribute in the at least one attribute are spliced to obtain the original vector corresponding to the data sample. With reference to the second aspect, in some embodiments, the second determining unit is specifically configured to: if the target category is a discrete numeric type, use a dictionary encoding algorithm to convert the attribute information corresponding to the attribute into the attribute corresponding to the attribute the representation vector of ; If the target category is a continuous numerical type, a numerical algorithm is used to convert the attribute information corresponding to the attribute into a representation vector corresponding to the attribute; If the target category is a text type or an image type, the Bert encoding algorithm is used to convert the attribute information corresponding to the attribute into a representation vector corresponding to the attribute. (here, LIU teaches that the vector embodies the type that it represents, if the type is ALL these types, as in the present embodiment, then the vector taught by LIU would be all these types.))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with LIU‘s attribute vectors and, with LEGRAND‘s document data, with a reasonable expectation of success, a method for attribute vectors of the data type of a document used for training, as in LIU, where the document data comes from documents selected by a user, as found in LEGRAND. A person of ordinary skill would have been motivated to generate a more effective vector space (LIU, Background paragraph 1).
While LEGRAND, in view of LIU, does teach attribute vectors used for training a model, it does not explicitly teach:
combining the respective attribute vectors of the sequence to generate a single sequence vector;
However, in analogous art that similarly uses vectors made of attributes, SONG teaches:
combining the respective attribute vectors of the sequence to generate a single sequence vector;
([0247] Accordingly, the computer device may combine the word vectors of the acquired text to be quantized to form a larger vector (i.e., the word vector sequence) as the input layer variable of the text vector prediction model of the first language text.)
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with SONG‘s vector combination method and, with LEGRAND‘s, as modified by LIU, document data, with a reasonable expectation of success, a method for vectors combined and inputted, as in SONG, where the vectors are made of document metadata, as found in LEGRAND, as modified by LIU. A person of ordinary skill would have been motivated to improve semantic analysis (SONG, [0007]).
LIU further teaches:
inputting the single sequence vector into a machine learning model; (( Page 3, line 3) Inputting the target output vector into the risk control model to be trained)
While LEGRAND, in view of LIU and SONG, does teach attribute vectors used for training a model, it does not explicitly teach:
and outputting, to the user, in response to the sequence of selections, a predicted next document selection according to an output of the machine learning model,
However, in analogous art that similarly tracks user selected documents, ZHOU teaches:
and outputting, to the user, in response to the sequence of selections, a predicted next document selection according to an output of the machine learning model based on the single sequence vector. ([0056] In a first aspect of the present disclosure, a method for ranking documents in search results is provided. The method includes defining a first training data set, the first training data set including, for each of a plurality of user queries, information respective of a document selected by a user from results responsive to the query and information respective of one or more documents within an observation window after the selected document in the results, defining a second training data set, the second training data set including, for each of the plurality of user queries, information respective of the selected document, training a first machine learning model with the first training data set, the first machine learning model configured to output a predicted user document selection, training a second machine learning model with the second training data set, the second machine learning model configured to output a predicted user document selection, and ranking documents of a further search result set according to the output of the first machine learning model and the output of the second machine learning model. (the predicted next selection is shown in the form of the ranking of the document, which is found using a learning model))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with ZHOU‘s document selection prediction and, with LEGRAND‘s, as modified by LIU and SONG, document data, with a reasonable expectation of success, a method for predicted the next selected document, as in ZHOU, where the selection is found using document data vectors, as found in LEGRAND, as modified by LIU and SONG. A person of ordinary skill would have been motivated to improve prediction accuracy (ZHOU, [0009]).
LIU further teaches:
Wherein generating the representative attribute vector comprises concatenating the numerical attribute vector portion, the category attribute vector portion, the text content vector portion, and the image content vector portion into the respective attribute vector, (LIU page 5, paragraph 12)When converting the target attribute information into the corresponding original vector, it is possible to determine the target category to which the attribute information corresponding to each attribute in the target attribute information belongs, and use the information conversion algorithm corresponding to the target category to convert the attribute information corresponding to the attribute to the attribute The corresponding representation vector is further spliced with the representation vector corresponding to each attribute, thereby obtaining the original vector corresponding to the data sample.)
While LEGRAND, in view of LIU, SONG, and ZHOU, does teach attribute vectors used for training a model, it does not explicitly teach:
wherein the plurality of vector portions used to generate the respective attribute vector of each document does not include a unique identifier of that document.
However, in analogous art that similarly handles document data, ZHANG teaches:
wherein the attribute vector respective of each document does not include any portion representative of a unique identifier of that document. ([0076]… The sentence vectors for the sentences in the data sources 218 can be referred to as the trained sentence vectors. The trained sentence vectors can be stored in the sentence vector database 212 in association with an identifier of which electronic document the trained sentence vector corresponds. For example, all the sentences from a first electronic document can be converted into sentence vectors and stored in the sentence vector database 212 in association with a unique identifier that identifies the first electronic document. (as seen here, the vectors are associated with the identifier, meaning that the vectors themselves do not include the identifiers))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with ZHANG‘s vectors not including identifiers and, with LEGRAND‘s, as modified by LIU, SONG, and ZHOU, document data, with a reasonable expectation of success, a method for vectors without document identifiers, as in ZHANG, where the vectors are made of document metadata, as found in LEGRAND, as modified by LIU, SONG, and ZHOU. A person of ordinary skill would have been motivated to improve search accuracy (ZHANG, [0002]).
Regarding claim 3, SONG further teaches:
The method of claim 1, wherein inputting the attribute vectors of the sequence into the machine learning model comprises inputting the single sequence vector into the machine learning model. ([0247] Accordingly, the computer device may combine the word vectors of the acquired text to be quantized to form a larger vector (i.e., the word vector sequence) as the input layer variable of the text vector prediction model of the first language text.)
Regarding claim 4, LEGRAND further teaches
The method of claim 1, further comprising: training the machine learning model according to a training data set, ([0389] In operation 2616 an embedding is learned in dependence upon only the provided training data, i.e. not based on any documents that are considered to be outside the subject domain. A machine learning algorithm can be used to learn this embedding.) the training data set comprising: a plurality of attribute vectors representative of a plurality of user document selection sequences, ([0048] In an implementation of the present disclosure, a system can have several aspects, and different implementation need not implement all of the following aspects: 1) a module for creating an initial query, 2) a module for obtaining a set of candidate results satisfying the initial query, 3) a module for determining the distance or similarity between candidate results or a module for embedding the candidate results in a vector space, [0075] In order to embed a document catalog in a vector space each document must be associated with a vector. A distance between two documents in such a space is then determined using standard measures of distance using vectors.)
LIU further teaches:
each attribute vector comprising: one or more numerical attribute vector portions representative of numerical attributes of one or more documents in one of the sequences; one or more category attribute vector portions representative of category information of one or more documents in one of the sequences; and one or more text content vector portions representative of text content of one or more documents in one of the sequences. one or more image vector portions representative of one or more images in one or more documents in one of the sequences. (Optionally, the attribute information to the target information of the conversion algorithm, the attribute corresponding to the attribute is information into the attribute corresponding to the representation vector, if the target category is a discrete value type, then can be dictionary for coding representation, such as one-hot code, obtaining the corresponding representation vector. if it is a continuous numerical value type, it can use the data after numerical value to represent, obtaining the corresponding representation vector, the representation vector is a one-dimensional vector. if it is text or image type, then using Bert code to encode the text or image, obtaining the corresponding representation vector.)
Regarding claim 5, ZHOU further teaches:
The method of claim 1, wherein outputting the predicted next document selection comprises one or more of: displaying a link to the predicted next document in response to a user search; displaying a link to the predicted next document in response to a user navigation; or displaying a link to the predicted next document in response to a user click. ([0013] The training data source 102 may include multiple types of training data. For example, in some embodiments, the training data source 102 may include biased data 110 and unbiased data 112. Both biased training data 110 and unbiased training data 112 may include a plurality of search queries, the search engine results for each query, and user behavior with respect to those results, including document selections (e.g., user clicks on document links)…. Accordingly, the unbiased data 112 may include the same search results sets and corresponding user behaviors as the biased data 110, but may include only the document links that are at and above the clicked link. [0062] In an embodiment of the first aspect, the method further includes displaying the ranked further search result set to a user.(the clickable results of a search engine are clickable links/urls to the documents))
Regarding claim 10, it comprises of limitations similar to those of claim 1 and is therefore rejected for similar rationale. Regarding claims 12-14, they comprise of limitations similar to those of claims 3-5 and are therefore rejected for similar rationale.
Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1) ], LIU (C.N. Pub. No. CN 114548273 A), SONG (U.S. Pub. No. US 20170060855 A1), ZHOU (U.S. Pub. No. US 20220138209 A1) in further view of MASUMOTO (U.S. Pub. No. US 20070177762 A1)
Regarding claim 6, while LEGRAND, as modified by LIU and ZHOU does teach claim 1, which claim 6 is dependent upon, it does not explicitly teach:
The method of claim 1, wherein the output of the machine learning model comprises a respective unique identifier of one or more predicted next documents.
However, in analogous art that similarly makes predictions, MASUMOTO teaches:
The method of claim 1, wherein the output of the machine learning model comprises a respective unique identifier of one or more predicted next documents. ([0146] The embedded data generating unit 1003 generates a document ID. The document ID is a unique ID and is generated for each document including a plurality of pages. When a plurality of copies of a document including a plurality of pages are to be output, a unique ID is generated for each copy. )
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with MASUMOTO‘s ID generation and, with LEGRAND‘s, as modified by LIU and ZHOU, document data, with a reasonable expectation of success, a method for generating an id for predicted documents, as in MASUMOTO, where the data used for the prediction comes from document metadata, as found in LEGRAND, as modified by LIU and ZHOU. A person of ordinary skill would have been motivated to prevent falsification (MASUMOTO, [0009]).
Regarding claim 15, it comprises of limitations similar to those of claim 6 and is therefore rejected for similar rationale.
Claims 7 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1) ], LIU (C.N. Pub. No. CN 114548273 A), SONG (U.S. Pub. No. US 20170060855 A1), ZHOU (U.S. Pub. No. US 20220138209 A1), MASUMOTO (U.S. Pub. No. US 20070177762 A1) in further view of GHOSH (U.S. Pub. No. US 20230359408 A1)
Regarding claim 7, while LEGRAND, as modified by LIU, ZHOU, and MASUMOTO does teach claim 6, which claim 7 is dependent upon, it does not explicitly teach:
The method of claim 6, wherein the output of the machine learning model further comprises a respective category of one or more predicted next documents.
However, in analogous art that similarly makes document predictions, GHOSH teaches:
The method of claim 6, wherein the output of the machine learning model further comprises a respective category of one or more predicted next documents. ([0024] Thus, machine learning model 108 may learn to predict the category of the document from the labelled training data. Example training of machine learning model 108 may be described in FIG. 5.)
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with GHOSH‘s category prediction and, with LEGRAND‘s, as modified by LIU, ZHOU, and MASUMOTO, document data, with a reasonable expectation of success, a method for predicting a category for the document, as in GHOSH, for the predicted document using the document metadata, as found in LEGRAND, as modified by LIU, ZHOU, and MASUMOTO. A person of ordinary skill would have been motivated to increase security (GHOSH, [0014]).
Regarding claim 16, it comprises of limitations similar to those of claim 7, and is therefore rejected for similar rationale.
Claims 8 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1) ], LIU (C.N. Pub. No. CN 114548273 A), ZHOU (U.S. Pub. No. US 20220138209 A1), MASUMOTO (U.S. Pub. No. US 20070177762 A1), GHOSH (U.S. Pub. No. US 20230359408 A1) in further view of LINDNER (U.S. Pub. No. US 20060026217 A1)
Regarding claim 8, while LEGRAND, as modified by LIU, ZHOU, MASUMOTO, and GHOSH does teach claim 7, which claim 8 is dependent upon, it does not explicitly teach:
The method of claim 7, wherein the respective categories of the one or more predicted next documents is used by the machine learning model to generate the respective unique identifiers of the one or more predicted next documents.
However, in analogous art that similarly deals with internet media, LINDNER teaches:
The method of claim 7, wherein the respective categories of the one or more predicted next documents is used by the machine learning model to generate the respective unique identifiers of the one or more predicted next documents. ([0021] For all other source storage media, the operator again scans the storage media's barcode (50) and enters the storage media's inspection metadata into the database (55). Entry of the inspection metadata into the database allows the operator to place the source storage media into discrete categories (60). Based on these categories, the database generates a Unique Material Identifier (UMID))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with LINDNER‘s ID generation and, with LEGRAND‘s, as modified by LIU, ZHOU, MASUMOTO, and GHOSH, document data and category prediction, with a reasonable expectation of success, a method for generating an id for a document, as in LINDNER, using its predicted category, as found in LEGRAND, as modified by LIU, ZHOU, MASUMOTO, and GHOSH. A person of ordinary skill would have been motivated to better handle long term data preservation (LINDNER, [0005]).
Regarding claim 17, it comprises of limitations similar to those of claim 8 and is therefore rejected for similar rationale.
Claims 9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1) ], LIU (C.N. Pub. No. CN 114548273 A), SONG (U.S. Pub. No. US 20170060855 A1), ZHOU (U.S. Pub. No. US 20220138209 A1) in further view of DANIELSON (U.S. Pub. No. US 20080082521 A1)
While LEGRAND, as modified by LIU and ZHOU, teaches claim 1, which claim 9 is dependent upon, it does not explicitly teach:
The method of claim 1, wherein the respective attribute vector for each document in the sequence comprises a unique identifier portion representative of a unique identifier of the document.
However, in analogous art that similarly uses attribute vectors, DANIELSON teaches:
The method of claim 1, wherein the respective attribute vector for each document in the sequence comprises a unique identifier portion representative of a unique identifier of the document. ([0026]… The document and attribute vector are then combined along with the unique document identifier to produce the default processing vector.)
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with DANIELSON‘s inclusion of a unique ID and, with LEGRAND‘s, as modified by LIU and ZHOU, attribute vector, with a reasonable expectation of success, a method for including the unique ID into the vector, as in DANIELSON, where the vector is made of document attributes, as found in LEGRAND, as modified by LIU and ZHOU. A person of ordinary skill would have been motivated to increase usability (DANIELSON, [0012]).
Regarding claim 18, it comprises of limitations similar to those of claim 9 and is therefore rejected for similar rationale.
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1) in view of ZHANG (U.S. Pub. No. US 20190221204 A1) in view of ZHOU (U.S. Pub. No. US 20220138209 A1)
Regarding claim 19, LEGRAND substantially teaches the claim including:
A method comprising:receiving, from a user, a sequence of selections of documents; ([0261] The user model can be designed and implemented to determine a probability that a document D would be chosen, given a set of documents presented to the user and the sequence of selections/clicks up to that point. (the sequence of clicks would be a sequence of selections the users made on documents)) generating, for each document in the sequence, a respective attribute vector, according to metadata respective of the document, ([0048] In an implementation of the present disclosure, a system can have several aspects, and different implementation need not implement all of the following aspects: 1) a module for creating an initial query, 2) a module for obtaining a set of candidate results satisfying the initial query, 3) a module for determining the distance or similarity between candidate results or a module for embedding the candidate results in a vector space, [0075] In order to embed a document catalog in a vector space each document must be associated with a vector. A distance between two documents in such a space is then determined using standard measures of distance using vectors. [0081] Often embeddings will consider documents in multiple ways. For example, a product may be embedded in terms of the meta-data associated with that product, the image of that product, and the textual content of reviews for that product. Such an embedding may be achieved by developing Kernels for each aspect of the document and combining those Kernels in some way, e.g., via a linear combination. (here, we see that the vector space, and the vectors embedded within, are embedded off of the metadata of the document selected))
While LEGRAND does teach getting an attribute vector from user selections, it does not explicitly teach:
but not according to a unique identifier of the document; inputting the attribute vectors of the sequence into a machine learning model;
However, in analogous art that similarly teaches using attribute vectors, ZHANG teaches:
but not according to a unique identifier of the document; inputting the attribute vectors of the sequence into a machine learning model; ([0076]… The sentence vectors for the sentences in the data sources 218 can be referred to as the trained sentence vectors. The trained sentence vectors can be stored in the sentence vector database 212 in association with an identifier of which electronic document the trained sentence vector corresponds. For example, all the sentences from a first electronic document can be converted into sentence vectors and stored in the sentence vector database 212 in association with a unique identifier that identifies the first electronic document. [0061] The word vector generator 206 can train the neural network using the training pairs generated by the vector generator 204. For example, for each training pair, the input term is one-hot encoded and used as an input to the neural network and the output terms are one-hot encoded and used as outputs of the neural network for the given input term. (as seen here, the vectors are associated with the identifier, meaning that the vectors themselves do not include the identifiers, further, the training pair is generated by a vector generator, meaning they are vectors.))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with ZHANG‘s vector training and, with LEGRAND‘s document data, with a reasonable expectation of success, a method for training a model with vectors that do not include unique IDs, as in ZHANG, where the document data and vectors come from documents selected by a user, as found in LEGRAND. A person of ordinary skill would have been motivated to improve search accuracy (ZHANG, [0002]).
While LEGRAND, as modified by ZHANG, does teach training a model using attribute vectors based on user selections, it does not explicitly teach:
and outputting, to the user, in response to the sequence of selections, a predicted next document selection according to an output of the machine learning model.
However, in analogous art that similarly teaches using a predictive model, ZHOU teaches:
and outputting, to the user, in response to the sequence of selections, a predicted next document selection according to an output of the machine learning model. ([0056] In a first aspect of the present disclosure, a method for ranking documents in search results is provided. The method includes defining a first training data set, the first training data set including, for each of a plurality of user queries, information respective of a document selected by a user from results responsive to the query and information respective of one or more documents within an observation window after the selected document in the results, defining a second training data set, the second training data set including, for each of the plurality of user queries, information respective of the selected document, training a first machine learning model with the first training data set, the first machine learning model configured to output a predicted user document selection, training a second machine learning model with the second training data set, the second machine learning model configured to output a predicted user document selection, and ranking documents of a further search result set according to the output of the first machine learning model and the output of the second machine learning model. (the predicted next selection is shown in the form of the ranking of the document, which is found using a learning model))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with ZHOU‘s document selection prediction and, with LEGRAND‘s, as modified by ZHANG, document data, with a reasonable expectation of success, a method for predicted the next selected document, as in ZHOU, where the selection is found using document data vectors, as found in LEGRAND, as modified by ZHANG. A person of ordinary skill would have been motivated to improve prediction accuracy (ZHOU, [0009]).
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over LEGRAND (U.S. Pub. No. US 20170091319 A1), ZHANG (U.S. Pub. No. US 20190221204 A1), ZHOU (U.S. Pub. No. US 20220138209 A1) in further view of LIU (C.N. Pub. No. CN 114548273 A)
Regarding claim 20, while LEGRAND, as modified by ZHANG and ZHOU, teaches claim 19, which claim 20 is dependent upon, it does not explicitly teach:
The method of claim 19, wherein generating a respective attribute vector for each document in the sequence is further according to a content of the document.
However, in analogous art that similarly uses attribute vectors, LIU teaches:
The method of claim 19, wherein generating a respective attribute vector for each document in the sequence is further according to a content of the document. ((LIU claim 5) The method according to claim 4, wherein the converting the attribute information corresponding to the attribute into a representation vector corresponding to the attribute using an information conversion algorithm corresponding to the target category, comprising: If the target category is a discrete numerical type, a dictionary encoding algorithm is used to convert the attribute information corresponding to the attribute into a representation vector corresponding to the attribute; If the target category is a continuous numerical type, a numerical algorithm is used to convert the attribute information corresponding to the attribute into a representation vector corresponding to the attribute; If the target category is a text type or an image type, the Bert encoding algorithm is used to convert the attribute information corresponding to the attribute into a representation vector corresponding to the attribute (here, LIU teaches that the vector embodies the type that it represents, if the type is ALL these types, as in the present embodiment, then the vector taught by LIU would be all these types.))
It would have been obvious to a person skilled in the art before the effective filing date of the invention to have combined with LIU‘s attribute vectors and, with LEGRAND‘s, as modified by ZHANG and ZHOU, document data, with a reasonable expectation of success, a method for attribute vectors of the data type of a document used for training, as in LIU, where the document data comes from documents selected by a user, as found in LEGRAND, as modified by ZHANG and ZHOU. A person of ordinary skill would have been motivated to generate a more effective vector space (LIU, Background paragraph 1).
Response to Arguments
Applicant’s arguments filed 22-JULY-2026 have been fully considered, but they are found to be non-persuasive
With regards to the applicant’s remarks regarding the 103 rejection in the non-final action, the applicant argues that the prior art does not teach the newly amended claims 1, and 10. The examiner acknowledges this argument and has adjusted the prior art of LEGRAND, LIU and ZHOU to disclose the newly added limitations. Namely, the newly added limitation regarding combining the portions together has been mapped to with LIU teaching splicing the representation vectors of the attributes together to make the original vector.
Further regarding the applicant’s remarks regarding the LIU reference of claim 1, the applicant argues:
Notwithstanding the above, even though Liu's "representation vector 1" and Liu's "representation vector 2" may be a plurality of vectors representative of the "data sample" of Liu, Liu does not teach "generating, for each document in the sequence, a respective attribute vector including a plurality of vector portions," and does not teach "wherein generating the respective attribute vector comprises concatenating the numerical attribute vector portion, the category attribute vector portion, the text content vector portion, and the image content vector portion into the respective attribute vector," as required by amended Claim 1, because the "representation vector 1" and the "representation vector 2" of Liu do not include the "numerical attribute vector portion," the "category attribute vector portion," the "text content vector portion," and the "image content vector portion," as required by amended Claim 1. Instead, Liu teaches that the "target category" can be the different types including the "discrete numerical type," the "continuous numerical type," or the "text or image type," which determines the algorithm that is used to convert the attribute information into the corresponding "attribute vector" of Liu.
With regards to this argument, this seems to be a misunderstanding of LIU’s wording. LIU does not teach that the portions are either a numerical, categorical, text, or image vector portion, but rather is explaining how it approaches the handling of vector portions of those types. LIU is instead both capable and used to convert all of those types for any give input. Further, LIU page 5, paragraph 12 states that LIU specifically does “include, but is not limited to, text, numbers, images, and so on.”
Even assuming, arguendo, that each "representation vector" of Liu is converted from the "attribute information" using an algorithm selected from one of the different types of "target category," which Applicant does not concede, Liu does not teach that the "original vector" that is representative of the original data sample includes the "numerical attribute vector portion," the "category attribute vector portion," the "text content vector portion," and the "image content vector portion," as required by amended Claim 1. Instead, Liu teaches that the "representation vector[s]" are spliced into the "original vector," and that each "representation vector" is based on the "target category" of the corresponding attribute information. That is, Liu teaches that each "representation vector" can be the "discrete numerical type," the "continuous numerical type," or the "text or image type," but Liu does not teach that the "original vector" includes a plurality of "representation vector[s]" including the "discrete numerical type," the "continuous numerical type," and the "text [type] and [the] image type," as required by amended Claim 1.
With regards to this argument, as stated above, LIU specifically does include each of the types listed in claim 1 and as such the representation vector, which is made for each attribute information, also would include each type.
Furthermore, Liu does not teach that the "representation vector" can be the "category attribute vector portion representative of category information," as required by amended Claim 1. Instead, each "representation vector" of Liu is converted from the attribute information based on the type of the "target category" of the attribute information. Accordingly, the "representation vector" of Liu is not representative of the "category information," but rather the "representation vector" of Liu is representative of the "attribute information." Therefore, the "representation vector" of Liu does not represent the type of "target category." Thus, Liu does not teach "generating, for each document in the sequence, a respective attribute vector including a plurality of vector portions," and does not teach "wherein generating the respective attribute vector comprises concatenating the numerical attribute vector portion, the category attribute vector portion, the text content vector portion, and the image content vector portion into the respective attribute vector," as recited in amended Claim 1.
With regards to this argument, the conversion from the attribute information into the representation vector does occur based on the attribute information’s category, however, the representation vector conversion is directly related to the category. This is because in LIU, the information of the attribute is directly correlated to the information of the category as each attribute is one of the categories as well. As such, the representation vector is a representation of that portion’s category and information. Therefore, the representative vector is both an attribute and category information vector. Further, neither of the relevant cited parts of claim 1 cite a ‘category attribute vector portion representative of category information’ and as such, even if it weren’t the case that LIU does teach one, LIU could still be used to teach the newly amended parts of claim 1.
For reasons similar to those stated in the foregoing, Song does not teach "combining the respective attribute vectors of the sequence of selections of documents to generate a single sequence vector," as recited in amended Claim 1, because Song's "word vectors" do not correspond to the "respective attribute vectors," as required by amended Claim 1. Instead, Song's "word vectors" are representative of the words of the acquired text. Song does not teach that the "larger vector" or the received text is representative of the "sequence of selections of documents," as required by amended Claim 1. Even though the acquired text is a sequence of words, Song does not teach that the "larger vector" is representative of a sequence of acquired texts. Accordingly, Song's "word vectors" are similar to Liu's "representation vector[s]," and therefore Song's "word vectors" are not representative of the "sequence of selections of documents," as required by amended Claim 1. Therefore, Song does not teach "combining the respective attribute vectors of the sequence of selections of documents to generate a single sequence vector," as recited in amended Claim 1.
Regarding this argument, while SONG does not specifically claim that its word vectors are representation vector portions of attributes, SONG does however teach the act of combining vector sequences to make a single sequence vector. A ‘attribute vectors of the sequence’ has already been taught by the combination of LIU and LEGRAND and as such, under U.S.C. 103, SONG can be used in combination with LIU and LEGRAND to teach the limitation in its entirety. One of ordinary skill in the art could use SONG’s methodology in combination with the vector portions as taught by LIU to achieve the claimed limitation.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SKIELER A KOWALIK whose telephone number is (571)272-1850. The examiner can normally be reached 8-5.
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, Mariela D Reyes can be reached at (571)270-1006. 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.
/SKIELER ALEXANDER KOWALIK/Examiner, Art Unit 2142
/HAIMEI JIANG/Primary Examiner, Art Unit 2142
Prosecution Insights