DETAILED ACTION
This Office Action is responsive to amendments and arguments filed on June 23rd, 2026. Claims 1, 6-7, 10, 12, 14-18 and 20 are amended, claims 1-20 are pending and have been examined; hence, this action is made FINAL.
Any objections/rejections not mentioned in this Office Action have been withdrawn by the Examiner.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
This application continues from prior Application No. 18/185,547, filed March 17th, 2023, and as such, is granted the benefit of the earlier filing date.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on October 6 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Response to Amendments and Arguments
Regarding rejections made under 35 U.S.C. 103, Applicant argues that “Mao only disclosed using heuristic rules or pattern based rules, different from ‘learning, using a machine learning model, a plurality of metric weights corresponding to a plurality of relationships types’ as recited in amended claim 1… Mao only mentions ‘the metadata can include structured information relating to entities’, but claim 1 recites ‘wherein the plurality of relationship types comprises at least two relationship types from a set including a position type, a hierarchical structure type, a style type, and a semantic type’,” (page 12 of Remarks) and provides figures and arguments to distinguish Mao’s teachings from the amended claims.
Applicant further argues a technical distinction, “Dernoncourt[‘s] ‘digital annotations’ refer to selections (or de-selections) of key portions of a document (mouse-click or drag and drop)… The loss function computed by comparing the predicted digital annotations to the generated ground-truth digital annotations is different from ‘… by computing a loss function based on a difference between a predicted anchor and a ground-truth anchor’ recited in claim 1,” (page 17 of Remarks).
Applicant’s arguments are moot, as the broadest reasonable interpretation of the amended claim language is taught by Damodaran in combination with U.S. Patent Application Publication 2024/0249543 to Jayaram et al. under 35 U.S.C. 103. Further details are provided below.
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, 3-11 and 13-20 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication 2021/0264208 to Damodaran (hereinafter, "Damodaran") in view of U.S. Patent Application Publication 2024/0249543 to Jayaram et al. (hereinafter, “Jayaram”).
Regarding claims 1, 10 and 15, Damodaran teaches a method, computer-readable medium and system comprising: obtaining a plurality of documents and a query, wherein the query indicates a common information element located in each of the plurality of documents (paragraph [0024], "The information server 202 is configured to receive instruction from the client device 204 for extract information from one or more documents. For example, the client device 204 can provide the information server 202 with a copy of a company's annual report, and the information server 202 can analyze the annual report document to extract fields (such as, chief executive officer (CEO), company name, total assets, etc.).";
identifying a plurality of flexible anchor elements corresponding to the common information element, wherein each of the plurality of flexible anchor elements has a different relationship to the common information element (paragraph [0042], "At step 308, the information server 202 can determine a first set of fields of interest for the first document. That is, at step 306, if the information server 202 determines that the first document is of a certain type, then the fields of interest associated with the certain type of document are used as the first set of fields of interest. For example, if the document type is determined to be an annual report, then the first fields of interest can include 'CEO', 'Total Liabilities', 'Total Assets', etc.");
learning, using a machine learning model, a plurality of metric weights corresponding to a plurality of relationship types, wherein the plurality of relationship types comprises at least two relationship types from a set including a position type, a hierarchical structure type, a style type, and a semantic type (paragraphs [0036]-[0039], "Fields of interest can have various properties that can be configured as depicted in FIG. 4. For example, 'CEO' field of interest configuration in an annual report can be linked to keywords of synonyms which include 'chief executive officer', 'group chief executive', and 'CEO' as shown in item 404. The 'CEO' field of interest configuration can also include the field name 402, the field origin 426 of where to look for values for the field, the field source 424 which can be a text source, and the field type 422 which indicates that the CEO field is a field of interest. The field name 402 merely names the field. The field origin 426 can take on values including paragraph, line, page, etc… Additional properties include a page affinity 412 which indicates a page of a document where the field is likely to be found… Additional properties include a field affinity 408 which indicates that some fields can be related.");
extracting, using the machine learning model, the common information element from each of the plurality of documents by applying the plurality of metric weights to the plurality of flexible anchor elements (paragraphs [0043]-[0045], "At step 310, the information server 202 determines which CDQA models provide answers to each field of interest in the first set of fields of interest. The information server 202 uses the CDQA models to ask specific questions. In some implementations, the questions are based on configurable properties of a field of interest. For example, if the field of interest is 'CEO', then a configuration of the field of interest can include 'Who is the . . . ?' as a type of question for probing the first document.
In some implementations, each of the CDQA models in the CDQA model repository 208 is used to probe the first document for each field of interest in the first set of fields of interest. CDQA models provide basic confidence scores along with potential answers for each field of interest.
In some implementations, the potential answers for each field of interest is further probed by the field of interest extractor 216 for veracity... For example, company names can be checked against a list of known companies, and depending on the result, the basic confidence score can be boosted. In another example, total liabilities is supposed to return a number, as such, if a CDQA model returns a non-numeric potential answer, then the basic confidence score associated with that non-numeric potential answer is reduced. After evaluating each of the potential answers from the different CDQA models, the answers with the highest confidence scores for each field of interest are determined as the best answers to the first set of fields of interest."); and
generating, in response to the query, content including values of the common information element from the plurality of documents (paragraph [0046], "At step 312, the information server 202 provides the best answers to the client device 204. The best answers provided can be curated as previously discussed in connection with the post processing engine 218 of FIG. 2. For example, the best answers can be provided in sentence format, in a tabular format, in a list format, etc.").
Damodaran does not explicitly disclose a method wherein “the machine learning model is trained to adapt the plurality of metric weights by computing a loss function based on a difference between a predicted anchor and a ground-truth anchor,” and thus, Jayaram is introduced.
Jayaram teaches a method and system for data extraction wherein the machine learning model is trained to adapt the plurality of metric weights by computing a loss function based on a difference between a predicted anchor and a ground-truth anchor (paragraphs [0078]-[0079], "As mentioned, in some examples, the first machine learning model is a feature-specific model which is exclusively trained for extraction associated with the first feature, and the second machine learning model is a feature-specific model which is exclusively trained for extraction associated with the second feature. The two machine learning models may thus be trained to analyze and predict values for different features. Training may involve supervised learning, e.g., using sample documents (with annotations) uploaded by a user via the web interface 130 of FIG. 1. In some examples, the first training set differs from the second training set. For example, the first training set may include annotations to provide 'ground truth' data related to the first feature, while the second training set may include annotations to provide 'ground truth' data related to the second feature. In some examples, one or more of the models may be validated with an annotated (labelled) training set as part of the supervised learning process.
The training process may thus include, for one or more of the sample documents in a training set, (a) annotating and/or correcting data related to the feature (e.g., field) that needs to be trained, and (b) saving/confirming the 'ground truth' for the one or more of the sample documents. The document information extraction system 122 may provide, via a suitable user interface, an annotation tool to facilitate annotation and/or correction during or prior to a training job.").
Damodaran and Jayaram are considered analogous because they are each concerned with adaptable data extraction. Given that the substitution of one known element for another would yield predictable results, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to have replaced the domain training of the CDQA models taught by Damodaran with the ground-truth training as taught by Jayaram for the purpose of improving model extraction performance, as taught by Damodaran.
Regarding claim 3, Damodaran teaches the method of claim 1, wherein: the content is generated in-situ within a software application that extracts a plurality of common information elements (paragraph [0024], "The information server 202 can include a document classification engine 212, a document pre-preprocessing engine 214, a field of interest extractor 216, and a post processor 218. An engine is a combination of hardware and software configured to perform specific functionality. The information server 202 is configured to receive instructions from the client device 204 for extracting information from one or more documents. For example, the client device 204 can provide the information server 202 with a copy of a company's annual report, and the information server 202 can analyze the annual report document to extract fields (such as, chief executive officer (CEO), company name, total assets, etc.). In some implementations, the information server 202 does not have to know the type of document being examined. Each of the document classification engine 212, the document processing engine 214, the field of interest extractor 216, and the post processing engine 218 identified in FIG. 2 is a combination of hardware and software configured to perform specific functionality as described in the following paragraphs."). In the teachings of Damodaran, the user interfaces with a device and the document processing and reporting are executed in a server environment. No teachings of Damodaran preclude the server processes from being conducted by a single software application, and from the perspective of the user, the request and response may be through the same interface.
Regarding claim 4, Damodaran teaches the method of claim 1, wherein: the content is generated by a generative machine learning model (paragraph [0020], "Embodiments of the present disclosure provide a zero-shot task transfer based domain agnostic information extraction framework. A plurality of closed-domain question answering (CDQA) models are leveraged to achieve domain agnostic document extraction. CDQA models are models trained under a specific domain.").
Regarding claim 5, Damodaran teaches the method of claim 1, further comprising: generating an additional document corresponding to the plurality of documents based on the generated content (paragraph [0031], "In some implementations, the post processing engine 218 provides the best answers in statement format to the client device 204 in an email, a chatbox, a voice recording, etc.").
Regarding claim 6, Damodaran teaches the method of claim 1, further comprising: generating an insight from the generated content, and presenting the insight to a user (paragraph [0030], "The post processing engine 218 cleanses the best answers for each field of interest for displaying on the client device 204. The post processing engine 218 can determine a best form for returning the best answers to the client device 204.").
Regarding claim 7, Damodaran teaches the method of claim 1, wherein: each of the plurality of flexible anchor elements comprises a plurality of relationships to a corresponding common information element of a plurality of common information elements (paragraph [0036], "Fields of interest can have various properties that can be configured as depicted in FIG. 4. For example, 'CEO' field of interest configuration in an annual report can be linked to keywords of synonyms which include 'chief executive officer', 'group chief executive', and 'CEO' as shown in item 404. The 'CEO' field of interest configuration can also include the field name 402, the field origin 426 of where to look for values for the field, the field source 424 which can be a text source, and the field type 422 which indicates that the CEO field is a field of interest. The field name 402 merely names the field. The field origin 426 can take on values including paragraph, line, page, etc.").
Regarding claim 8, Damodaran teaches the method of claim 1, wherein: the query is a natural language query interpreted by a natural language processing (NLP) model (paragraph [0018], "Machine learning approaches to document extraction also present a challenge when trying to sidestep the previously described training problem by combining different machine learning models. For example, a classification model can be used to determine domain, then a named-entity-recognition (NER) model can be used for synonyms of fields of interest, and then a natural language query can be used to extract information pertaining to the fields of interest.").
Regarding claim 9, Damodaran teaches the method of claim 8, wherein: the NLP model is a Bidirectional Encoder Representations from Transformers (BERT) or Generative pre-trained transformers (GPT) model (paragraph [0029], "The field of interest extractor 216 utilizes at least two CDQA models when extracting fields from the document. The document elements (e.g., lines, paragraphs, etc.) determined by the document preprocessing engine 214 are candidates for probing with the at least two CDQA models. Examples of CDQA models include Bidirectional Encoder Representations from Transformers (BERT) trained on Stanford Question Answering Dataset (SQuAD), Simple Bi-Directional Attention Flow (BiDAF), ELMo-BIDAF, etc.").
Regarding claim 11, Damodaran teaches the non-transitory computer-readable medium of claim 10, wherein the executable instructions further cause the processing device to perform operations comprising: presenting a generated insight to a user based on the content (paragraph [0030], "The post processing engine 218 cleanses the best answers for each field of interest for displaying on the client device 204. The post processing engine 218 can determine a best form for returning the best answers to the client device 204," and paragraph [0031], "In some implementations, the post processing engine 218 provides the best answers in statement format to the client device 204 in an email, a chatbox, a voice recording, etc.").
Regarding claim 13, Damodaran teaches the non-transitory computer-readable medium of claim 10, wherein the executable instructions further cause the processing device to perform operations comprising: receiving a natural language query from a user, and identifying one or more common information elements in at least one of the plurality of documents based on the query (paragraph [0018], "Machine learning approaches to document extraction also present a challenge when trying to sidestep the previously described training problem by combining different machine learning models. For example, a classification model can be used to determine domain, then a named-entity-recognition (NER) model can be used for synonyms of fields of interest, and then a natural language query can be used to extract information pertaining to the fields of interest.").
Regarding claim 14, Damodaran teaches the non-transitory computer-readable medium of claim 13, wherein the executable instructions further cause the processing device to perform operations comprising: parsing the natural language query using a Bidirectional Encoder Representations from Transformers (BERT) or Generative pre-trained transformers (GPT) model (paragraph [0029], "The field of interest extractor 216 utilizes at least two CDQA models when extracting fields from the document. The document elements (e.g., lines, paragraphs, etc.) determined by the document preprocessing engine 214 are candidates for probing with the at least two CDQA models. Examples of CDQA models include Bidirectional Encoder Representations from Transformers (BERT) trained on Stanford Question Answering Dataset (SQuAD), Simple Bi-Directional Attention Flow (BiDAF), ELMo-BIDAF, etc. The field of interest extractor 216 asks questions on the document elements of the preprocessed document using the different CDQA models in order to determine which of the available CDQA models provides a best response for a given question.").
Regarding claim 16, Damodaran teaches the system of claim 15, further comprising: clustering the plurality of documents to obtain a document cluster; and obtaining a plurality of common information elements from the document cluster (paragraph [0032], "In some implementations, the post processing engine 218 can link different documents together that may be related. For example, in a contracts example, a master service agreement, a statement of work, and an addendum can be preprocessed and elements indexed by the preprocessing engine 214. The field of interest extractor 216 can extract fields like a 'First Party' field, a 'Second Party' field, and a 'Master Service Agreement effective date' field. The post processing engine 218 can link the master service agreement, the statement of work, and the addendum in a knowledge graph if the 'First Party' field, the 'Second Party' field, and the 'Master Service Agreement effective date' field match in the three documents.").
Regarding claim 17, Damodaran teaches the system of claim 15, wherein: the plurality of documents includes unstructured documents (paragraph [0002], "The present disclosure relates to information extraction from documents and more specifically to systems and methods for extracting information of interest from unfamiliar documents using zero-shot learning," and paragraph [0003], "Information extraction involves automatically extracting structured information from either structured or unstructured documents. In some cases, the unstructured documents are human language texts where natural language processing (NLP) is applied to extract the structured information.").
Regarding claim 18, Damodaran teaches the system of claim 15, further comprising: receiving the query from a user (paragraph [0018], "Machine learning approaches to document extraction also present a challenge when trying to sidestep the previously described training problem by combining different machine learning models. For example, a classification model can be used to determine domain, then a named-entity-recognition (NER) model can be used for synonyms of fields of interest, and then a natural language query can be used to extract information pertaining to the fields of interest.");analyzing the query using a natural language processor (paragraph [0003], "Information extraction involves automatically extracting structured information from either structured or unstructured documents. In some cases, the unstructured documents are human language texts where natural language processing (NLP) is applied to extract the structured information."); andgenerating an insight in response to the query based on the analysis (paragraph [0030], "The post processing engine 218 cleanses the best answers for each field of interest for displaying on the client device 204. The post processing engine 218 can determine a best form for returning the best answers to the client device 204.").
Regarding claim 19, Damodaran teaches the system of claim 18, further comprising: an analysis model trained to automatically perform the analysis and generate the insight (paragraph [0024], "The information server 202 is configured to receive instructions from the client device 204 for extracting information from one or more documents. For example, the client device 204 can provide the information server 202 with a copy of a company's annual report, and the information server 202 can analyze the annual report document to extract fields (such as, chief executive officer (CEO), company name, total assets, etc.). In some implementations, the information server 202 does not have to know the type of document being examined.").
Regarding claim 20, Damodaran teaches the system of claim 19, wherein: the common information element is identified using the machine learning model trained to identify a plurality of common information element (paragraph [0020], "Embodiments of the present disclosure provide a zero-shot task transfer based domain agnostic information extraction framework. A plurality of closed-domain question answering (CDQA) models are leveraged to achieve domain agnostic document extraction. CDQA models are models trained under a specific domain," and paragraph [0026], "The field of interest extractor 216 can extract fields within the document using different CDQA models and using different document extraction configurations. The document extraction configurations can include extraction configurations for annual reports, invoices, statements of work, master service agreements, etc.").
Claims 2 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Damodaran and Jayaram as applied to claims 1 and 10 above, further in view of U.S. Patent Application Publication 2003/0212664 to Breining et al. (hereinafter, "Breining").
Regarding claim 2, the combination of Damodaran and Jayaram does not explicitly teach “The method of claim 1, wherein the content comprises a table containing a plurality of data elements,” and thus, Breining is introduced.
Breining teaches the content comprises a table containing a plurality of data elements (paragraph [0043], "In the manner described above, the XML data is parsed into a single flat table and queried using conventional techniques to produce the output 60.").
Damodaran, Jayaram and Breining are considered analogous because they are each concerned with identifying and extracting data. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified Damodaran and Jayaram with the teachings of Breining for the purpose of improving data readability. Given that all the claimed elements were known in the prior art, one skilled in the art could have combined the elements by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results.
Regarding claim 12, the combination of Damodaran and Jayaram does not explicitly teach “The non-transitory computer-readable medium of claim 10, wherein the executable instructions further cause the processing device to perform operations comprising: inputting the one or more extracted common information elements into a flat table having a denormalized schema,” however, Breining teaches inputting the one or more extracted common information elements into a flat table having a denormalized schema (paragraph [0043], "In the manner described above, the XML data is parsed into a single flat table and queried using conventional techniques to produce the output 60. The method described next avoids having to flatten the XML data into a single table and thereby repeat information in that table. Instead, the method operates on the XML information "on-the-fly," as a wrapper extracts it from the XML document."). In paragraph [0058], Applicant indicates that, while duplication is permitted in denormalized tables, the claimed invention may avoid duplicating data entries. The teachings of Breining also indicate the creation of a flat representation of data, where duplication is permitted and therefore the schema denormalized, but with efforts to avoid data duplication.
Damodaran, Jayaram and Breining are considered analogous because they are each concerned with identifying and extracting data. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified Damodaran and Jayaram with the teachings of Breining for the purpose of improving data readability. Given that all the claimed elements were known in the prior art, one skilled in the art could have combined the elements by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results.
Conclusion
THIS ACTION IS MADE FINAL. 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.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
U.S. Patent 7,428,699 to Kane et al.
U.S. Patent 7,644,361 to Wu et al.
U.S. Patent 7,840,564 to Holzgrafe et al.
U.S. Patent 8,285,074 to Saund et al.
U.S. Patent 8,832,655 to Grechanik.
U.S. Patent 8,954,438 to Mao et al.
U.S. Patent 10,628,525 to Fink et al.
U.S. Patent 11,200,073 to Voicu.
U.S. Patent Application Publication 2012/0124086 to Song et al.
U.S. Patent Application Publication 2015/0106156 to Chang et al.
U.S. Patent Application Publication 2018/0025436 to Déjean.
U.S. Patent Application Publication 2018/010725 to Schmidt.
U.S. Patent Application Publication 2019/0384807 to Dernoncourt et al.
U.S. Patent Application Publication 2020/0104350 to Allen et al.
U.S. Patent Application Publication 2020/0160050 to Bhotika et al.
U.S. Patent Application Publication 2021/0248153 to Sirangimoorthy et al.
U.S. Patent Application Publication 2022/0043858 to Rinehart et al.
U.S. Patent Application Publication 2022/0188509 to Zeng et al.
U.S. Patent Application Publication 2022/0207268 to Gligan et al.
U.S. Patent Application Publication 2023/0169813 to Muthu et al.
U.S. Patent Application Publication 2024/0303412 to Evans et al.
“One-shot text field labeling using attention and belief propagation for structure information extraction” by Cheng et al.
"PDF-to-Text Reanalysis for Linguistic Data Mining” by Goodman et al.
“Facilitating conversational interaction in natural language interfaces for visualization” by Mitra et al.
“NL4DV: A Toolkit for Generating Analytic Specifications for Data Visualization from Natural Language Queries” by Narechania et al.
“Landmarks and Regions: A Robust Approach to Data Extraction” by Parthasarathy et al.
“Form2Seq: A Framework for Higher-Order Form Structure Extraction” by Aggarwal et al.
“DocStruct: A Multimodal Method to Extract Hierarchy Structure in document for General Form Understanding” by Want et al.
“Web Data Extraction Using Textual Anchors” by Pouramini et al.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEAN T SMITH whose telephone number is (571)272-6643. The examiner can normally be reached Monday - Friday 8:00am - 5:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, PIERRE-LOUIS DESIR can be reached at (571) 272-7799. 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.
/SEAN THOMAS SMITH/Examiner, Art Unit 2659
/PIERRE LOUIS DESIR/Supervisory Patent Examiner, Art Unit 2659
Prosecution Insights