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
Applicant’s Remarks, filed January 1st, 2026, has been fully considered and entered. Accordingly, claims 1-12 are pending in the case. Claim 1 was amended. Claim 1 is the independent claim.
In light of Applicant’s Amendment, the 35 USC 101 Rejection of claims 1-12 has been withdrawn.
In light of Applicant’s Amendment, the 35 USC 112 Rejection of claim 9 has been withdrawn.
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 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 set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied 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 1-3, and 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Paruchuri et al. (US 2022/0156300 A1) in view of Arvela et al. (US 2022/0004545 A1), further in view of Mane et al. (US 2022/0100797 A1).
Regarding claim 1, Paruchuri teaches a system for graph-based clustering of documents, the system comprises one or more processors configured to:
receive a digital copies of a plurality of documents, each document is converted into a graph object (see Paruchuri, Paragraphs [0017], [0019], “A document processing system configured to process structured and unstructured documents with handwritten and printed inputs for entity extraction is disclosed. One or more documents that are accessed are initially processed for image generation so that each image corresponds to one of the documents. In an example, the documents can be received as scanned images.” [The documents can be received as scanned images (i.e., digital copies).]);
determine and label entities in each document, wherein each of the entities is represented as a node of the graph object; create the graph object for the received digital copy of each document (see Paruchuri, Paragraphs [0017], [0019], “The masked documents are converted into images so that each image includes a textual unit or a token with one or more words. … A graphical representation of the document to be processed by the trained visual language model is initially generated. The graphical representation includes interconnected nodes wherein each node (e.g., a word) is connected to adjacent nodes with edges. The trained visual-language model processes the deep document data structure to set the weights from the text, position, and image data which is then concatenated with the node embeddings from the graphical representation of the document to generate predictions for one or more of the names/values of the entities in the tokens.” [A graph representation (i.e., graph object) including nodes, and edges is created. The nodes may be a word (i.e., entity).]);
However, Paruchuri does not explicitly teach:
generate a graph embedding vector using a graph embedding neural network trained to receive the graph object as input and generate the graph embedding vector for the graph object as output;
Arvela teaches:
generate a graph embedding vector using a graph embedding neural network trained to receive the graph object as input and generate the graph embedding vector for the graph object as output (see Arvela, Paragraph [0070], “The engine 16 comprises a graph embedding engine that converts graphs into multidimensional vector format using a model trained by a graph embedding trainer 14A of the neural network trainer 14 using reference data from the document reference data store 10C” [The graph embedding engine (i.e., graph embedding neural network) can convert the graphs into vectors (i.e., graph embedding vectors).]);
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined Paruchuri (teaching deep document processing with self-supervised learning) in view of Arvela (teaching method of searching patent documents), and arrived at a system that incorporates a graph embedding neural network. One of ordinary skill in the art would have been motivated to make such a combination for the purposes of efficiently searching texts (see Arvela, Paragraph [0009]). In addition, both the references (Paruchuri and Arvela) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as data processing. The close relation between both the references highly suggests an expectation of success.
However, the combination of Paruchuri, and Arvela do not explicitly teach:
cluster the graph embedding vectors into a plurality of clusters, the plurality of clusters including at least a first cluster corresponding to a first template and a second cluster corresponding to a second template, wherein the first template is structurally different from the second template;
Mane teaches:
cluster the graph embedding vectors into a plurality of clusters, the plurality of clusters including at least a first cluster corresponding to a first template and a second cluster corresponding to a second template, wherein the first template is structurally different from the second template (see Mane, Paragraph [0034], “As part of clustering in step 232, the feature vectors in the list of feature vectors are partitioned into a selected number of clusters. … The documents in the document set DS are also clustered according to the clustering of their corresponding feature vectors. Each cluster is associated with a respective layout template, and the knowledge of that layout can be used for extraction of information from the document(s) belonging to the cluster.” [The vectors are clustered based on a layout template (i.e., first template and second template, wherein the first template is structurally different from the second template).]);
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined Paruchuri (teaching deep document processing with self-supervised learning) in view of Arvela (teaching method of searching patent documents), further in view of Mane (teaching system and method for clustering documents), and arrived at a system that incorporates clustering based on a type of structure. One of ordinary skill in the art would have been motivated to make such a combination for the purposes of improving classification of documents based on document layout structure (see Mane, Paragraph [0005]). In addition, the references (Paruchuri, Arvela, and Mane) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as data processing. The close relation between the references highly suggests an expectation of success.
The combination of Paruchuri, and Arvela further teaches:
and assign each document of the plurality of documents to the first template or the second template based on similarity between the graph embedding vector of the document and graph embedding vectors corresponding to the plurality of clusters, wherein each of the first template and the second template comprises documents having similar looking templates (see Mane, Paragraph [0034], “As part of clustering in step 232, the feature vectors in the list of feature vectors are partitioned into a selected number of clusters. … The documents in the document set DS are also clustered according to the clustering of their corresponding feature vectors. Each cluster is associated with a respective layout template, and the knowledge of that layout can be used for extraction of information from the document(s) belonging to the cluster.” [The documents are clustered based on the layout template.]).
Regarding claim 2, Paruchuri in view of Arvela, further in view of Mane teaches all the limitations of claim 1. Paruchuri further teaches:
connecting each of the nodes representing an entity with its neighbouring nodes along four directions; and forming edges between each of the nodes and its neighbouring nodes along four directions (see Paruchuri, Paragraphs [0036], [0044], “each of the tokens obtained from an unstructured document using the selected OCR tool can be treated as a node with edges connecting the node to adjacent nodes so that each token (or textual unit) can be connected to its four nearest neighbors (e.g,, top, bottom, left, right). … Each of the textual units generated from the unstructured documents in the data set by the selected OCR tool can be treated as being connected to a minimum number of neighboring nodes (e.g., four neighboring nodes) via edges.” [A node representing a word (i.e., entity) may be connected to its four nearest neighbors, which forms edges between the node and its four neighboring nodes.]).
Regarding claim 3, Paruchuri in view of Arvela, further in view of Mane teaches all the limitations of claim 2. Paruchuri further teaches:
wherein the edges are formed between the nodes based on the relative position of each of the nodes with its neighbouring nodes (see Paruchuri, Paragraphs [0036], [0044], “each of the tokens obtained from an unstructured document using the selected OCR tool can be treated as a node with edges connecting the node to adjacent nodes so that each token (or textual unit) can be connected to its four nearest neighbors (e.g,, top, bottom, left, right). … Each of the textual units generated from the unstructured documents in the data set by the selected OCR tool can be treated as being connected to a minimum number of neighboring nodes (e.g., four neighboring nodes) via edges.” [The edges with the four neighboring nodes are formed relative to the node.]).
Regarding claim 11, Paruchuri in view of Arvela, further in view of Mane teaches all the limitations of claim 1. Mane further teaches:
cluster the graph embedding vectors using clustering techniques such as partitional clustering such as K-means clustering, hierarchical clustering such as agglomerative clustering, or spectral clustering (see Mane, Paragraph [0033], “clustering of the feature vectors is performed in step 232 using one or more known clustering techniques. Examples of such clustering techniques include, but are not limited to k-means clustering, mean-shift clustering, expectation-maximization (EM) clustering using Gaussian mixture models (GMM), agglomerative hierarchical clustering, etc.” [Clustering techniques such as K-means clustering, and agglomerative clustering may be used.]).
Regarding claim 12, Paruchuri in view of Arvela, further in view of Mane teaches all the limitations of claim 1. Arvela further teaches:
the system comprises a machine learning model configured to classify the documents, wherein the clustered documents are fed as input to the machine learning model (see Arvela, Paragraph [0073], “User data obtained from the user interface 18 is fed after embedding in the embedding unit 13 to the graph embedding engine for vectorization, after which a vector comparison engine 16B finds a set of closest vectors corresponding to the graphs of the graph store 10B. The set of closest graphs is fed to graph classifier engine 16C, which compares them one by one with the user graph, using the trained graph classifier model in order to get accurate matches.” [The set of closest vectors (i.e., cluster of graph embedding vectors) are fed as input to trained graph classifier model (i.e., machine learning model) in order to get accurate matches (i.e., classify the documents).]).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Paruchuri in view of Arvela in view of Mane, further in view of Wang et al. (US 2023/0342558 A1).
Regarding claim 4, Paruchuri and Arvela, further in view of Mane teaches all the limitations of claim 1. However, the combination of Paruchuri, Arvela, and Mane do not explicitly teach:
wherein the graph embedding neural network is a Siamese network comprising: a first neural network comprising: a first encoder; a first graph neural network; and a first pooling layer; and a second neural network comprising: a second encoder; a second graph neural network; and a second pooling layer.
Wang teaches:
wherein the graph embedding neural network is a Siamese network comprising: a first neural network comprising: a first encoder; a first graph neural network; and a first pooling layer; and a second neural network comprising: a second encoder; a second graph neural network; and a second pooling layer (see Wang, Paragraph [0078], “FIGS. 6A-6B show exemplary networks of layers of machine learning model used by generalized entity matching system (GEM) 100 of FIG. 1 , consistent with embodiments of the present disclosure. GEM 100 may use a machine learning (ML) model including layers: input layer 121, encoding layer 122, pooling layer 123, and output layer 124. GEM 100 may use different network architectures such as sequenced architecture (as shown in FIG. 6A) and siamese architecture (as shown in FIG. 6B) to identify matched entity pairs.” [A Siamese network may be implemented, which includes first and second encoders and pooling layers.]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined Paruchuri (teaching deep document processing with self-supervised learning) in view of Arvela (teaching method of searching patent documents) in view of Mane (teaching system and method for clustering documents), further in view of Wang (teaching systems and methods for generalized entity matching), and arrived at a system that incorporates a Siamese network. One of ordinary skill in the art would have been motivated to make such a combination for the purposes of finding matches efficiently (see Wang, Paragraph [0079]). In addition, the references (Paruchuri, Arvela, Mane, and Wang) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as data processing. The close relation between the references highly suggests an expectation of success.
Claims 5-10 are rejected under 35 U.S.C. 103 as being unpatentable over Paruchuri in view of Arvela in view of Mane in view of Wang, further in view of Amin et al. (US 2024/0126988 A1).
Regarding claim 5, Paruchuri and Arvela in view of Mane, further in view of Wang teaches all the limitations of claim 4. Arvela further teaches:
train the graph embedding neural network using a training dataset comprising training documents, wherein graph embedding neural network is trained by: identifying and labelling entities, using the processor, in each of the training documents, wherein each of the entities is represented as a node of a graph object; creating graph objects, using the processor, for each of the training documents (see Arvela, Paragraph [0070], “The engine 16 comprises a graph embedding engine that converts graphs into multidimensional vector format using a model trained by a graph embedding trainer 14A of the neural network trainer 14 using reference data from the document reference data store 10C” [The graph embedding neural network may be trained.]);
However, the combination of Paruchuri, Arvela, Mane, and Wang do not explicitly teach:
computing graph edit distance (GED) matrix, using the processor, for a first batch of documents from the training dataset;
Amin teaches:
computing graph edit distance (GED) matrix, using the processor, for a first batch of documents from the training dataset (see Amin, Paragraphs [0075], [0108], “This lexical representation may be a table, a matrix, a graph, or the like, for example. … the lexical representation generation unit 422 calculates the normalized edit distance of each node in the first lexical representation with respect to each node in the second lexical representation. Normalized edit distance is a measure of sentence similarity and may be calculated by Equation 1 as shown below.” [A graph edit distance may calculated in order to determine similarity.]);
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined Paruchuri (teaching deep document processing with self-supervised learning) in view of Arvela (teaching method of searching patent documents) in view of Mane (teaching system and method for clustering documents) in view of Wang (teaching systems and methods for generalized entity matching), further in view of Amin (teaching a word extraction device, system and method), and arrived at a system that incorporates a graph edit distance technique. One of ordinary skill in the art would have been motivated to make such a combination for the purposes determining similarity (see Amin, Paragraph [0108]). In addition, the references (Paruchuri, Arvela, Mane, Wang, and Amin) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as data processing. The close relation between the references highly suggests an expectation of success.
The combination of Paruchuri, Arvela, Mane, Wang, and Amin further teaches:
inputting a pair of graph objects and the computed graph edit distance matrix to the graph embedding neural network, wherein one of the graph objects is input to the first neural network and the other graph object is input to the second neural network; generating a graph embedding vector, by the graph embedding neural network, for each of the input pair of graph objects; and calculating a similarity score between the graph embedding vectors generated by the first neural network and the second neural network (see Arvela, Paragraphs [0070], [0077], “The engine 16 comprises a graph embedding engine that converts graphs into multidimensional vector format using a model trained by a graph embedding trainer 14A of the neural network trainer 14 using reference data from the document reference data store 10C … the graph classifier engine 16C, as trained by the graph classifier trainer 14C, outputs similarity scores, which are the higher the more similar the compared graphs are in terms of both node content and nodal structure, as learned from the reference data using a learning target dependent thereof.” [The graph embedding neural network can convert the graphs into vectors (i.e., graph embedding vectors). Similarity scores may be determined.] Also, see Wang, Paragraph [0078], [A Siamese network may be implemented.] Also, see Amin, Paragraph [0108], “the lexical representation generation unit 422 calculates the normalized edit distance of each node in the first lexical representation with respect to each node in the second lexical representation. Normalized edit distance is a measure of sentence similarity and may be calculated by Equation 1 as shown below.” [A graph edit distance may calculated in order to determine similarity.]).
Regarding claim 6, Paruchuri in view of Arvela in view of Mane in view of Wang, further in view of Amin teaches all the limitations of claim 5. Arvela, and Wang further teaches:
wherein: the first encoder and the second encoder are configured to receive the graph object as input and generate an entity type embedding vector of a predefined size; the first graph neural network and the second graph neural network are configured to generate node representations encoding the structural information of the documents of the graph objects; and the first pooling layer and the second pooling layer are configured to aggregate the node representations and generate the graph embedding vector for the input graph object (see Arvela, Paragraph [0070], “The engine 16 comprises a graph embedding engine that converts graphs into multidimensional vector format using a model trained by a graph embedding trainer 14A of the neural network trainer 14 using reference data from the document reference data store 10C” [The graph embedding engine can convert the graphs into vectors (i.e., graph embedding vectors).] Also, see Wang, Paragraph [0078], “FIGS. 6A-6B show exemplary networks of layers of machine learning model used by generalized entity matching system (GEM) 100 of FIG. 1 , consistent with embodiments of the present disclosure. GEM 100 may use a machine learning (ML) model including layers: input layer 121, encoding layer 122, pooling layer 123, and output layer 124. GEM 100 may use different network architectures such as sequenced architecture (as shown in FIG. 6A) and siamese architecture (as shown in FIG. 6B) to identify matched entity pairs.” [A Siamese network is implemented, which includes first and second encoders and pooling layers.]).
Regarding claim 7, Paruchuri in view of Arvela in view of Mane in view of Wang, further in view of Amin teaches all the limitations of claim 5. Amin further teaches:
normalize the computed graph edit distance of the graph edit distance matrix to be between the range of 0 to 1 (see Amin, Paragraphs [0075], [0112], “This lexical representation may be a table, a matrix, a graph, or the like, for example. … the lexical representation generation unit 422 may vary the normalization edit distance criterion in steps from “0.0” to “0.9” (the lower the normalization edit distance, the greater the similarity between the nodes),” [The graph edit matrix may be normalized between 0 to 1]).
Regarding claim 8, Paruchuri in view of Arvela in view of Mane in view of Wang, further in view of Amin teaches all the limitations of claim 5. Arvela further teaches:
wherein the graph embedding vector is vector of the size 1x128 (see Arvela, Paragraph [0071], “The graph embedding engine can convert the graphs into vectors having at least 100 dimensions, preferably 200 dimensions or more and even 300 dimensions or more.” [The graph embedding engine can convert the graphs into vectors (i.e., graph embedding vectors) that have many dimensions.]).
Regarding claim 9, Paruchuri in view of Arvela in view of Mane in view of Wang, further in view of Amin teaches all the limitations of claim 5. Arvela, and Wang further teaches:
wherein upon training the graph embedding neural network, only one neural network from the Siamese network is configured to generate the graph embedding vector for the input graph object (see Arvela, Paragraph [0070], “The engine 16 comprises a graph embedding engine that converts graphs into multidimensional vector format using a model trained by a graph embedding trainer 14A of the neural network trainer 14 using reference data from the document reference data store 10C” [The graph embedding neural network may be trained.] Also, see Wang, Paragraph [0078], “FIGS. 6A-6B show exemplary networks of layers of machine learning model used by generalized entity matching system (GEM) 100 of FIG. 1 , consistent with embodiments of the present disclosure. GEM 100 may use a machine learning (ML) model including layers: input layer 121, encoding layer 122, pooling layer 123, and output layer 124. GEM 100 may use different network architectures such as sequenced architecture (as shown in FIG. 6A) and siamese architecture (as shown in FIG. 6B) to identify matched entity pairs.” [A Siamese network may be implemented.]).
Regarding claim 10, Paruchuri in view of Arvela in view of Mane in view of Wang, further in view of Amin teaches all the limitations of claim 5. Arvela, and Amin further teaches:
wherein: the similarity score is calculated using a cosine similarity function; and the similarity score is represented as (1-GED) (see Arvela, Paragraphs [0077]-[0078], “the graph classifier engine 16C, as trained by the graph classifier trainer 14C, outputs similarity scores, which are the higher the more similar the compared graphs are in terms of both node content and nodal structure, as learned from the reference data using a learning target dependent thereof. Through training, the similarity scores of positive training cases (graphs depicting the same concept) derived from the reference data can be maximized, whereas the similarity scores of negative training cases (graphs depicting different concepts), are maximized. … Cosine similarity is one possible criterion for similarity of graphs or vectors derived therefrom.” [A cosine similarity function may be used to determine a similarity score.] Also, see Amin, Paragraph [0108], “the lexical representation generation unit 422 calculates the normalized edit distance of each node in the first lexical representation with respect to each node in the second lexical representation. Normalized edit distance is a measure of sentence similarity and may be calculated by Equation 1 as shown below.” [A graph edit distance may calculated in order to determine similarity.]).
Response to Arguments
Applicant’s Arguments, filed January 1st, 2026, have been fully considered, but are moot in light of the new grounds of rejection.
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 extension fee 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 date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUSAM TURKI SAMARA whose telephone number is (571)272-6803. The examiner can normally be reached on Monday - Thursday, Alternate Fridays.
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 on (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 an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/HUSAM TURKI SAMARA/
Examiner, Art Unit 2161
/APU M MOFIZ/Supervisory Patent Examiner, Art Unit 2161