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 .
Response to Amendment
The Amendment filed March 20, 2026, has been entered. Claims 1, 3 – 5, 7 – 11, 13 – 15 and 17 – 20 are pending in the application.
Response to Arguments
Applicant’s arguments, filed March 20, 2026, with respect to the 35 U.S.C. 112(b) rejection of claim 14 have been fully considered and are persuasive. The 35 U.S.C. 112(b) rejection of March 6, 2026, has been withdrawn.
Applicant’s arguments, filed March 20, 2026, with respect to the 35 U.S.C. 103 rejections of claims 1, 3 – 5, 7 – 11, 13 – 15 and 17 – 20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 103
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, 7 – 8, 11, 13 – 15, 17 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng (US Patent No. 10,482,118) in view of Mekala et al. ("SCDV: Sparse Composite Document Vectors using soft clustering over distributional representations"), hereinafter Mekala, and Lauber (US Patent No. 12,093,648).
Regarding claim 1, Zheng discloses a computer-implemented method for generating a fixed-size N-dimensional vector representation for a given document, the method comprising:
generating a plurality of N clusters C_1, C_2,..., C_N, from a plurality of documents, wherein each of the N clusters represents semantically similar text parts (Column 5, lines 35-40, "In some implementations, the word-vector engine 202 receives text data 212, and processes the text data 212 to provide word-vectors 214. In some examples, the text data 212 is provided from a corpus of documents, and can be collectively referred to as a vocabulary. An example corpus of documents can include Google News, or Wikipedia."; Column 6, lines 34-44, "In some implementations, the grouping module 206 groups the words based on similarity scores of their respective word-vectors to provide groups (pseudo-words), referred to herein as clusters. Each cluster includes a plurality of words. Accordingly, the grouping module 206 provides pseudo-words (clusters) 216 as output. In some examples, the cosine similarity scores of word-vectors are compared to one another, and if the cosine similarity scores are determined to be sufficiently similar, the words represented by the respective word vectors are included in a cluster."; Column 8, lines 3-12, "Similarity scores are determined (306). For example, the similarity scoring module 204 receives the word-vectors 214, and determines similarity scores between the word-vectors 214 (e.g., as cosine similarity scores). Words are clustered (308). For example, unique words of the vocabulary are clustered into groups based on similarity scores of their respective word-vectors. In some examples, the grouping module 206 groups the words based on similarity scores of their respective word-vectors to provide groups (pseudo-words, clusters)."; Grouping words based on similarity scores of their respective word vectors to provide clusters reads on generating a plurality of N clusters, wherein each of the N clusters represents semantically similar text parts.),
and wherein the clustering is based on a relative distance of text-portion vectors to each other using a similarity function to determine distance (Column 6, lines 34-44, “In some implementations, the grouping module 206 groups the words based on similarity scores of their respective word-vectors to provide groups (pseudo-words), referred to herein as clusters. Each cluster includes a plurality of words. Accordingly, the grouping module 206 provides pseudo-words (clusters) 216 as output. In some examples, the cosine similarity scores of word-vectors are compared to one another, and if the cosine similarity scores are determined to be sufficiently similar, the words represented by the respective word vectors are included in a cluster.”; Including words in a cluster when the cosine similarity scores of word-vectors are compared to one another and the cosine similarity scores are determined to be sufficiently similar reads on the clustering being based on a relative distance of the text-portion vectors to each other using a similarity function to determine distance.);
generating an N-dimensional document vector E(D) for a document D by associating its nth coordinate value E(D)_n to the nth cluster C_n (Column 3, lines 2-5, "providing a document representation for each document in the plurality of documents, each document representation including a feature vector, each feature corresponding to a cluster."; Column 6, lines 26-36, "The document representation engine 210 receives the weighted clusters 218, and provides document representations for each document of the text data 212 based thereon. In some examples, each document representation is provided as a feature vector as similarly described above with reference to the BOW model. In the context of the present disclosure, the document representations could be described as being based on a bag-of-clusters (BOC) model, because each feature is provided as a cluster. In some examples, the dimension of the feature vector is equal to the number of clusters."; Generating document representations as a feature vector for each document based on the clusters, where the dimension of the feature vector is equal to the number of clusters and each feature corresponding to a cluster, reads on generating an N-dimensional document vector E(D) for a document D by associating its nth coordinate value to the nth cluster.);
processing further the N-dimensional document vector E(D) for one selected out of the group comprising document scoring, document classification, document similarity search, document similarity explanation, and document clustering (Column 3, line 63 - Column 4, line 1, "As introduced above, document classification can be performed using a machine-learning process, in which documents form a corpus of text that is used to train a machine-learning model. To perform such document classification, each document is processed to provide a respective document representation."; Column 5, lines 13-14, "In some examples, the document representations can be used to train a document classifier.");
extracting text-portions from the document D and embedding the text-portions extracted from the document D into K-dimensional text-portion vectors (Column 3, lines 2-5, "providing a document representation for each document in the plurality of documents, each document representation including a feature vector, each feature corresponding to a cluster."; Column 5, lines 43-48, "In some implementations, the word-vector engine 202 processes the text data 212 using Word2vec, which can be described as a group of related models that are used to produce word-vectors (also referred to as word embeddings). In some examples, each word-vector has multiple dimensions (e.g., hundreds of dimensions)."; Column 7, lines 26-28, "The document representation engine 210 receives the weighted clusters 218, and provides document representations for each document of the text data 212 based thereon."; Using Word2vec to generate word vectors where each word vector has multiple dimensions reads on extracting text-portions from the document D and embedding the extracted text-portions into K-dimensional text-portion vectors.);
and setting the values E(D)_n based on similarity matching score values between the text-portion vectors of the document D and text-portions vectors of the N clusters (Column 3, lines 2-5, "providing a document representation for each document in the plurality of documents, each document representation including a feature vector, each feature corresponding to a cluster."; Column 6, lines 56-64, "In some implementations, the pseudo-words 216 are input to the weighting module 208. The weighting module 208 provides a weight associated with each cluster. In accordance with implementations of the present disclosure, the weighting module 208 determines weights for the clusters using TF-IDF, and provides weighted clusters 218 as output. In some examples, TF-IDF can be described as a numerical statistic that conveys a relative importance of a word to a document in a collection of documents."; Column 7, lines 48-57, "In this example, the features of the vectors are the clusters that include words of the document in question, and the values are the respective TF-IDF weights. The number of dimensions is q (i.e., the number of clusters). If, however, all of the words of the document are not in a particular cluster, the value of that cluster is set equal to 0. In the depicted example, c1 and c4 include words of the document in question, hence include weight values, while c2, c3, c5, and cq do not include words of the document in question, hence include values of 0."; Generating document representations as a feature vector for each document based on the clusters, where the dimension of the feature vector is equal to the number of clusters and each feature corresponding to a cluster, and the feature vector values are the TF-IDF weights for the clusters, reads on setting the values E(D)_n based on similarity matching score values between the text-portion vectors of the document D and text-portions vectors of the N clusters, where the TF-IDF being a numerical statistic that conveys a relative importance of a word to a document reads on a similarity matching score value between the text-portion vectors of the document D and text-portions vectors of the N clusters.).
Zheng does not specifically disclose: wherein the plurality of clusters form a fixed-sized vocabulary.
Mekala teaches:
wherein the plurality of clusters form a fixed-sized vocabulary (Section 3.1, lines 1-9, “We begin by learning d dimensional word vector representations for every word in the vocabulary V using the skip-gram algorithm with negative sampling (SGNS) (Mikolov et al., 2013a). We then cluster these word embeddings using the Gaussian Mixture Models(GMM) (Reynolds, 2015) soft clustering technique. The number of clusters, K, to be formed is a parameter of the SCDV model.”; The number of clusters to be formed being a parameter reads on the plurality of clusters forming a fixed-sized vocabulary.).
Mekala is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng to incorporate the teachings of Mekala to cluster word embeddings where the number of clusters to be formed is a parameter. Doing so would allow for efficiently creating a more accurate semantic representation of documents (Mekala; Section 6, lines 1-18).
Zheng in view of Mekala does not specifically disclose: wherein the text-portion vectors comprise a mixture of different text types.
Lauber teaches:
wherein the text-portion vectors comprise a mixture of different text types (Column 4, line 55 - Column 5, line 1, "In one embodiment, an a priori or personalization score is itself based on two components: an inverse frequency score such as a SW score; and a topic relevancy score. A topic relevancy score for one of several topics associated with a document may be created from clusters. For example, once a phrase embedding model for a corpus has been trained, a clustering or grouping algorithm such as the K-Means algorithm or other algorithms can be used to cluster the embedding, of the phrases in a model (e.g. a system producing embeddings such as a table or a neural-network based model) into semantic topics. Clustering may include unsupervised machine learning techniques for grouping similar items (e.g. phrases, words, etc.) by some predefined distance metric."; Column 7, lines 29-46, "Analysis center 50 may perform functions such as those shown in FIG. 3, and may include for example embedding module 52 which may be or may be implemented as a machine learning or neural network algorithm, or by another system. Embedding module 52 may for example create embedding vectors or embedding model 54. Embedding model 54 may be a table including entries for tokens, words, phrases, etc., with associated embeddings created by embedding module 52, and associated weights or frequency scores. Embedding model 54 may be part of embedding module 52, or a separate unit. Analysis center 50 may communicate with for example user terminals to for example provide visualizations or the output of inference or production, conduct searches, etc. Embedding module 52 may be or includes a lookup-table directly mapping phrases, single tokens and/or multi-word terms to their associated embeddings, with keys of lookup table tokens and terms rather than structured phrases."; A phrase embedding model that clusters the embeddings, where the embedding module maps phrases, single tokens, and multi-word terms to their associated embeddings, reads on the text-portion vectors comprising a mixture of different text types.).
Lauber is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala to incorporate the teachings of Lauber to clusters embeddings where phrases, single tokens, and multi-word terms are mapped to their associated embeddings. Doing so would allow for generating a semantic representation of a text document to group together documents such as calls which are semantically similar (Lauber; Column 1, lines 54-62).
Regarding claim 3, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1.
Zheng further discloses:
wherein the plurality of documents is associated with a knowledge domain (Column 5, lines 35-40, "In some implementations, the word-vector engine 202 receives text data 212, and processes the text data 212 to provide word-vectors 214. In some examples, the text data 212 is provided from a corpus of documents, and can be collectively referred to as a vocabulary. An example corpus of documents can include Google News, or Wikipedia."; A corpus of document reads on the plurality of documents being associated with a knowledge domain.).
Regarding claim 4, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1.
Zheng further discloses:
further comprising: updating an E(D)_n value when a similarity matching score value between the text-portion vector of the document D and a text-portion of the nth cluster vector C_n is larger than similarity matching score values between the text-portion vector of the document D and the text-portion vectors from the N clusters excluding the nth cluster C_n (Column 6, lines 56-64, "In some implementations, the pseudo-words 216 are input to the weighting module 208. The weighting module 208 provides a weight associated with each cluster. In accordance with implementations of the present disclosure, the weighting module 208 determines weights for the clusters using TF-IDF, and provides weighted clusters 218 as output. In some examples, TF-IDF can be described as a numerical statistic that conveys a relative importance of a word to a document in a collection of documents."; Column 7, lines 48-57, "In this example, the features of the vectors are the clusters that include words of the document in question, and the values are the respective TF-IDF weights. The number of dimensions is q (i.e., the number of clusters). If, however, all of the words of the document are not in a particular cluster, the value of that cluster is set equal to 0. In the depicted example, c1 and c4 include words of the document in question, hence include weight values, while c2, c3, c5, and cq do not include words of the document in question, hence include values of 0."; Generating document representations as a feature vector for each document based on the clusters, where the dimension of the feature vector is equal to the number of clusters and each feature corresponding to a cluster, and the feature vector values are the TF-IDF weights for the clusters, reads on updating an E(D)_n value when a similarity matching score value between the text-portion vector of the document D and a text-portion of the nth cluster vector C_n is larger than similarity matching score values between the text-portion vector of the document D and the text-portion vectors from the N clusters excluding the nth cluster C_n, where the TF-IDF being a numerical statistic that conveys a relative importance of a word to a document and weights of zero being assigned for clusters that do not include the word reads on a similarity matching score value for cluster C_n being larger than the similarity matching score for the N clusters excluding the nth cluster C_n.).
Regarding claim 5, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1.
Lauber further teaches:
wherein the text-portions are multi-resolution text-portions and the clusters are multi-resolution clusters (Column 4, line 55 - Column 5, line 1, "In one embodiment, an a priori or personalization score is itself based on two components: an inverse frequency score such as a SW score; and a topic relevancy score. A topic relevancy score for one of several topics associated with a document may be created from clusters. For example, once a phrase embedding model for a corpus has been trained, a clustering or grouping algorithm such as the K-Means algorithm or other algorithms can be used to cluster the embedding, of the phrases in a model (e.g. a system producing embeddings such as a table or a neural-network based model) into semantic topics. Clustering may include unsupervised machine learning techniques for grouping similar items (e.g. phrases, words, etc.) by some predefined distance metric."; Column 7, lines 29-46, "Analysis center 50 may perform functions such as those shown in FIG. 3, and may include for example embedding module 52 which may be or may be implemented as a machine learning or neural network algorithm, or by another system. Embedding module 52 may for example create embedding vectors or embedding model 54. Embedding model 54 may be a table including entries for tokens, words, phrases, etc., with associated embeddings created by embedding module 52, and associated weights or frequency scores. Embedding model 54 may be part of embedding module 52, or a separate unit. Analysis center 50 may communicate with for example user terminals to for example provide visualizations or the output of inference or production, conduct searches, etc. Embedding module 52 may be or includes a lookup-table directly mapping phrases, single tokens and/or multi-word terms to their associated embeddings, with keys of lookup table tokens and terms rather than structured phrases."; A phrase embedding model that clusters the embeddings, where the embedding module maps phrases, single tokens, and multi-word terms to their associated embeddings, reads on the text-portions being multi-resolution text-portions and the clusters being multi-resolution clusters.).
Lauber is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala and Lauber to further incorporate the teachings of Lauber to clusters embeddings where phrases, single tokens, and multi-word terms are mapped to their associated embeddings. Doing so would allow for generating a semantic representation of a text document to group together documents such as calls which are semantically similar (Lauber; Column 1, lines 54-62).
Regarding claim 7, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1.
Lauber further teaches:
wherein the processing further is performed using a neural network system (Column 1, lines 54-62, "The semantic representation of a text document as a document embedding may be based on its components' embeddings. The question then becomes how to combine these into a single embedding which will represent the document as a whole. The possible reasons for doing this include the ability to group together documents such as calls which are semantically similar (e.g. using clustering), or searching for calls which are similar to a some query or indeed, to a specified call."; Column 17, lines 49-58, "Embodiments may improve on prior NLP, semantic and other technology by providing more accurate and meaningful document embeddings, which in turn may allow for more accurate and meaningful analysis of documents, e.g. improved searching over the documents, improved grouping or categorization of the documents, etc. Embedding module 52 may be, or may be implemented using, a NN, an information processing paradigm that may include nodes, referred to as neurons, organized into layers, with links between the neurons."; Grouping or categorization of the documents reads on document classification and document clustering, and the embedding module being implemented using a neural network (NN) reads on the processing being performed using a neural network system.).
Lauber is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala and Lauber to further incorporate the teachings of Lauber to generate a single embedding which represents a document as a whole and use the embedding to group or categorize documents, where the processing is performed using a neural network. Doing so would allow for generating a semantic representation of a text document to group together documents such as calls which are semantically similar (Lauber; Column 1, lines 54-62).
Regarding claim 8, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1.
Zheng further discloses:
wherein the document D is selected out of the plurality of documents (Column 3, lines 2-5, "providing a document representation for each document in the plurality of documents, each document representation including a feature vector, each feature corresponding to a cluster."; Providing a document representation for each document in a plurality of documents reads on the document D being selected out of the plurality of documents.).
Lauber further teaches:
wherein the document D is a new document (Column 2, line 60 - Column 3, line 1, "Embodiments of the present invention may determine an embedding (e.g. a numerical representation of the semantic content, such as a vector) for a document such as a call transcript or another text document. The embedding may be for a “newly seen” document, as a corpus of documents may be processed to create data, such as phrase embeddings and topics, to be used to determine the embedding of the new document: the new document may not be in the previously analyzed corpus."; Column 9, lines 5-11, "Operations 400-420 may be performed to process a corpus of documents, calls or interactions to create data such as embeddings that may be used to process documents, or new documents, and operations 430 onward may be performed for each document, or for a newly seen document (e.g. not in the corpus used to generate data), in order to create an embedding for that document."; Determining an embedding for a new document reads on the document D being a new document.).
Lauber is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala and Lauber to further incorporate the teachings of Lauber to determine an embedding for a new document that is not in a previously analyzed corpus. Doing so would allow for generating a semantic representation of a text document to group together documents such as calls which are semantically similar (Lauber; Column 1, lines 54-62).
Regarding claim 11, arguments analogous to claim 1 are applicable. In addition, Zheng discloses: a computer system (Column 8, lines 39-40, "Referring now to FIG. 4, a schematic diagram of an example computing system 400 is provided.") for generating a fixed-size N-dimensional vector representation for a given document, the computer system comprising: one or more computer processors, one or more computer-readable storage media, and program instructions stored on the one or more of the computer-readable storage media for execution by at least one of the one or more processors (Column 8, lines 51-53, "The processor 410 is capable of processing instructions stored in the memory 420 or on the storage device 430"), wherein the computer system is capable of performing a method comprising the steps of claim 1.
Regarding claim 13, arguments analogous to claim 3 are applicable.
Regarding claim 14, Zheng in view of Mekala and Lauber discloses the system as claimed in claim 11.
Zheng further discloses:
further comprising: updating an E(D)_n value when a similarity matching score value between a text-portion vector of the document D and a best matching text-portion vector from C_n is larger than similarity matching score values between the text-portion vector of the document D and the text- portion vectors from the N clusters excluding the nth cluster C_n (Column 6, lines 56-64, "In some implementations, the pseudo-words 216 are input to the weighting module 208. The weighting module 208 provides a weight associated with each cluster. In accordance with implementations of the present disclosure, the weighting module 208 determines weights for the clusters using TF-IDF, and provides weighted clusters 218 as output. In some examples, TF-IDF can be described as a numerical statistic that conveys a relative importance of a word to a document in a collection of documents."; Column 7, lines 48-57, "In this example, the features of the vectors are the clusters that include words of the document in question, and the values are the respective TF-IDF weights. The number of dimensions is q (i.e., the number of clusters). If, however, all of the words of the document are not in a particular cluster, the value of that cluster is set equal to 0. In the depicted example, c1 and c4 include words of the document in question, hence include weight values, while c2, c3, c5, and cq do not include words of the document in question, hence include values of 0."; Generating document representations as a feature vector for each document based on the clusters, where the dimension of the feature vector is equal to the number of clusters and each feature corresponding to a cluster, and the feature vector values are the TF-IDF weights for the clusters, reads on updating an E(D)_n value when a similarity matching score value between the text-portion vector of the document D and a best matching text-portion of the nth cluster vector C_n is larger than similarity matching score values between the text-portion vector of the document D and the text-portion vectors from the N clusters excluding the nth cluster C_n, where the TF-IDF being a numerical statistic that conveys a relative importance of a word to a document and weights of zero being assigned for clusters that do not include the word reads on a similarity matching score value for cluster C_n being larger than the similarity matching score for the N clusters excluding the nth cluster C_n.).
Regarding claim 15, arguments analogous to claim 5 are applicable.
Regarding claim 17, arguments analogous to claim 7 are applicable.
Regarding claim 20, arguments analogous to claim 1 are applicable. In addition, Zheng discloses: a computer program product (Column 8, lines 57-58, "In one implementation, the memory 420 is a computer-readable medium.") for generating a fixed-size vector representation for a given document, the computer program product comprising: one or more computer-readable storage media and program instructions stored on the one or more computer-readable storage media (Column 8, lines 51-53, "The processor 410 is capable of processing instructions stored in the memory 420 or on the storage device 430"), the program instructions executable by a computing system to cause the computing system to perform a method comprising the steps of claim 1.
Claims 9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng in view of Mekala and Lauber, and further in view of Bender et al. (US Patent No. 11,699,432), hereinafter Bender.
Regarding claim 9, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1, but does not specifically disclose: wherein the text-portion vectors of a cluster are represented by a centroid vector of the cluster.
Bender teaches:
wherein the text-portion vectors of a cluster are represented by a centroid vector of the cluster (Column 1, lines 42-51, "Some aspects include a process that includes obtaining a corpus of natural-language text documents. The process includes classifying the natural-language text documents into a plurality of classes corresponding to different fields of domain knowledge and obtaining a language model that maps n-grams contained in the natural-language text documents to one or more respective vectors in an embedding space, where at least some of the n-grams is mapped to a plurality of different vectors corresponding to different fields among the plurality of classes."; Column 17, lines 10-31, "Some embodiments may use an unsupervised learning operation to map one or more concepts represented by vertex groups to n-grams. For example, some embodiments may determine a vertex group using a clustering method, such as a K-means clustering method or a hierarchical clustering method, to determine a vector cluster. Each respective vector of the vector cluster corresponds with a respective vertex of the vertex group. Vertices of a vertex group may be described as being of the same cluster if their corresponding vectors are assigned to the same cluster during a clustering operation. For example, some embodiments may use a K-means clustering method after determining an initial set of centroids of vectors in a multi-sense embedding space. Some embodiments may determine the initial set of centroids based on an initial knowledge graph, determine a set of neighboring vertices of the centroid based on a set of pairwise distances between the set of neighboring vertices and the centroid in the embedding space, and re-compute each of the respective centroids based on the set of neighboring vertices. The use of the K-means clustering method may provide a fast way of determining groups of vertices and their associated n-grams."; Using a K-means clustering method after determining an initial set of centroids of vectors in a multi-sense embedding space reads on the vectors of a cluster being represented by a centroid vector of the cluster.).
Bender is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala and Lauber to incorporate the teachings of Bender to use a K-means clustering method after determining an initial set of centroids of vectors in a multi-sense embedding space. Doing so would allow for determining learned representations for documents to improve data ingestion accuracy when performing document analysis or text summarization operations (Bender; Column 4, lines 23-62).
Regarding claim 18, arguments analogous to claim 9 are applicable.
Claims 10 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng in view of Mekala and Lauber, and further in view of Croutwater et al. (US Patent Application Publication No. 2020/0110834), hereinafter Croutwater, and Sahayaraj et al. (US Patent Application Publication No. 2022/0180068), hereinafter Sahayaraj.
Regarding claim 10, Zheng in view of Mekala and Lauber discloses the method as claimed in claim 1, but does not specifically disclose: further comprising: upon changing a number of the plurality of documents, perform the following steps: adjusting the values of K and N; re-clustering the text-portion vectors of the plurality of documents; and re-generating the document vector of the document D.
Croutwater teaches:
upon changing a number of the plurality of documents, perform the following steps: adjusting the value of N (Paragraph 0036, lines 1-19, "It is understood that in linguistics, the corpus is a collection and structured set of texts. The corpus is itself subject to change. Such change may be based on removed a text from the corpus or adding a text to the corpus. Similarly, such change may be based on a change in the composition or content of a text that is a member of the corpus. As the corpus changes, the generated outcome (170) may also be subject to change. The linguistic manager (154) addresses the dynamic characteristic(s) of the corpus by application of a linguistic term to the cluster representation responsive to a detected change in the composition of the corpus. Similarly, the linguistic manager (154) is not limited to a single application of a linguistic term to the cluster representation. In one embodiment, the linguistic manager (154) may apply a sequence of linguistic terms to the cluster representation, thereby effectively training the cluster representation against the linguistic terms. Accordingly, the linguistic manager (154) is configured to adapt to the corpus and the associated cluster representation(s)."; Adapting the cluster representation to a change in the corpus, where a change in the corpus is removing a text from the corpus or adding a text to the corpus, reads on adjusting the value of N upon changing the number of the plurality of documents.);
re-clustering the text-portion vectors of the plurality of documents (Paragraph 0031, lines 16-23, "A clustering algorithm may be applied to the vector representation to find interesting data within the vector representation. Clustering is directed at grouping similar text units within a collection of documents. The document manager (154) forms one or more cluster representations of documents or files (162) and/or data resources (166a)-(166c), with each cluster representing a common topic."; Paragraph 0036, lines 2-11, "The corpus is itself subject to change. Such change may be based on removed a text from the corpus or adding a text to the corpus. Similarly, such change may be based on a change in the composition or content of a text that is a member of the corpus. As the corpus changes, the generated outcome (170) may also be subject to change. The linguistic manager (154) addresses the dynamic characteristic(s) of the corpus by application of a linguistic term to the cluster representation responsive to a detected change in the composition of the corpus."; Changing the cluster representation in response to a detected change in the composition of the corpus, where clustering is directed at grouping similar text units within a collection of documents with each cluster representing a common topic, reads on re-clustering the text-portion vectors of the plurality of documents.);
and re-generating the document vector of the document D (Paragraph 0026, lines 1-4, "Referring to FIG. 1, a schematic diagram of a computer system (100) is depicted to provide context to word vector and document vector representations, and linguistic processing responsive to the representations."; Paragraph 0031, lines 16-23, "A clustering algorithm may be applied to the vector representation to find interesting data within the vector representation. Clustering is directed at grouping similar text units within a collection of documents. The document manager (154) forms one or more cluster representations of documents or files (162) and/or data resources (166a)-(166c), with each cluster representing a common topic."; Paragraph 0036, lines 2-11, "The corpus is itself subject to change. Such change may be based on removed a text from the corpus or adding a text to the corpus. Similarly, such change may be based on a change in the composition or content of a text that is a member of the corpus. As the corpus changes, the generated outcome (170) may also be subject to change. The linguistic manager (154) addresses the dynamic characteristic(s) of the corpus by application of a linguistic term to the cluster representation responsive to a detected change in the composition of the corpus."; Changing the cluster representation in response to a detected change in the composition of the corpus and forming a cluster representation of a document reads on re-generating the document vector of the document D.).
Croutwater is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala and Lauber to incorporate the teachings of Croutwater to change the cluster representation of a collection of documents and form a document vector representation in response to a detected change in the composition of the corpus. Doing so would allow for implementing a text mining application that supports a query of a subset of documents and returns document content and statistical analysis of data associated with each facet (Croutwater; Paragraph 0025, lines 1-22).
Zheng in view of Mekala and Lauber, and further in view of Croutwater, does not specifically disclose: upon changing the number of the plurality of documents, perform the following steps: adjusting the value of K.
Sahayaraj teaches:
upon changing the number of the plurality of documents, adjusting the value of K (Paragraph 0033, lines 27-38, "The word embeddings may also reflect the size of the vocabulary of the respective text corpus or of the portion of the text corpus fed into the embedding model, because a vector may be kept for each word in the vocabulary of the text corpus that is fed in or of the text corpus portion that is fed in. This vocabulary size is separate from the dimensionality. For example, a word embedding for a large text corpus may have one hundred dimensions and may have one hundred thousand respective vectors for one hundred thousand unique words. The dimensions for the word embedding may relate to how each word in the text corpus relates to other words in the text corpus."; Changing the word embedding size to reflect the size of the vocabulary of a respective text corpus reads on adjusting the value of K upon changing the number of the plurality of documents.).
Sahayaraj is considered to be analogous to the claimed invention because it is in the same field of generating vector representations for documents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Zheng in view of Mekala and Lauber, and further in view of Croutwater, to incorporate the teachings of Sahayaraj to change the word embedding size to reflect the size of the vocabulary of a respective text corpus. Doing so would allow for the tracking of bias levels in a text corpus and the continual decreasing of bias over time in documents for a business or research community (Sahayaraj; Paragraph 0015, lines 1-21).
Regarding claim 19, arguments analogous to claim 10 are applicable.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to James Boggs whose telephone number is (571)272-2968. The examiner can normally be reached M-F 8:00 AM - 5:00 PM.
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, Daniel Washburn can be reached at (571)272-5551. 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.
/JAMES BOGGS/Examiner, Art Unit 2657
Prosecution Insights