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 .
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim(s) 1, 10-12, 14 and 17-18 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Document Classification and Information Extraction framework for Insurance Applications to GV et al., hereinafter, “GV”.
Claim 1. GV teaches A method comprising: [Abstract] Document Intelligence is an essential subclass in the field of machine learning… document classification and entity extraction
receiving, by a processor, an electronic document, the electronic document including text data; [Fig. 1] Document processing involves processing the document into page-level images.
extracting, by the processor and using text recognition, the text data from the electronic document; [II. Related Work] Information extraction from documents has traditionally been treated as a text-based task
[2) Character Based Sequence Classification] In this workflow, OCR is used to extract text from the document.
[C. Document Pre-processing] Images are then processed using PyTesseract OCR to extract text from images.
[E. Information Extraction result] Textual information is derived from OCR processing
inputting, by the processor, the text data to a machine-learning (ML) model, wherein the ML model is trained to receive text data as input and to output a document category that corresponds to the input; Fig. 1, Data Ingestion- Input (interpreted to be text data input), Document Processing-Stage 1 and Document Classification- Stage 2 (interpreted to be document category)
receiving, by the processor, as output from the ML model and based at least in part on the text data, an indication of a particular document category corresponding to the electronic document; Fig. 1, Document Classification and Prediction
[Introduction] This stage extracts relevant information based on the application. The final output is the prediction of both the classification and extraction model based on the business rules.
identifying, by the processor and based on the particular document category, a text segment of the text data, the text segment characterized by text features; Fig. 3 Ensemble VGG + Text Features: The final model we trained and currently use in production incorporates text-level features along with image features.
determining, by the processor and based on the text features, a data field of a claim processing system corresponding to the text segment; [2) Character Level Bidirectional LSTM Sequence Classification] This model tends to learn a better representation among fields - name, DOB, claim number, PID in the medical application, and claim id, year, model, make, and VIN in the salvage application… It also understands that specific fields are only numeric, only characters, or alpha-numeric. Since the model captures this information, there call for claim id, DOB and VIN is high compared to LayoutLM as these fields are often few characters in length and always have a generic pattern.
and associating, by the processor, the text segment with the data field. LayoutLM [13] is a pre-training method that combines the text of the document with the layout of the 2D document
[2) Character Level Bidirectional LSTM Sequence Classification] This model tends to learn a better representation among fields - name, DOB, claim number, PID in the medical application, and claim id, year, model, make, and VIN in the salvage application… It also understands that specific fields are only numeric, only characters, or alpha-numeric.
Claim 10. GV teaches The method of claim 1, wherein the document category comprises one of: a repair estimate, a repair bill, a medical bill, a medical treatment estimate, a vehicle rental bill, a tow bill, a police report or an attorney demand package. [Figure 9], [Introduction] These documents contain critical information such as personal identifiers, claim numbers, policy identifiers, medical providers, and insured property information that communicate across business workflows and systems.
Claim 11. GV teaches The method of claim 1, wherein the data field comprises one of: name of a policy participant, policy number, cause-of-loss code, name of a vendor, an address of the vendor, claim amount, estimated amount, vehicle make and model, or date of loss. [Figure 9], [B. Datasets and Experiment Setups] We labeled and annotated 700 salvage documents for extracting the relevant fields. Both the applications first classify these documents and then extract information: claim number, name, DOB, and patient id for medical bills; claim number, VIN, make, model, and year for salvage claims.
Claim 12. Reviewed and analyzed in the same way as claim 1. See the above analysis and rationale.
Claim 14. GV teaches The system of claim 12, wherein the text features comprise at least one of: a length of the text segment, whether the text segment is alphabetical or numerical, whether the text segment denotes a proper noun, whether the text segment denotes a calendar date, or whether the text segment denotes a mailing address. [2) Character Level Bidirectional LSTM Sequence Classification] This model tends to learn a better representation among fields - name, DOB, claim number, PID in the medical application, and claim id, year, model, make, and VIN in the salvage application… It also understands that specific fields are only numeric, only characters, or alpha-numeric.
Claim 17. Reviewed and analyzed in the same way as claim 1. See the above analysis and rationale.
Claim 18. Reviewed and analyzed in the same way as claim 14. See the above analysis and rationale.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 2 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Document Classification and Information Extraction framework for Insurance Applications to GV et al., hereinafter, “GV” and in view of US 2018/0144042 A1 to Sheng et al., hereinafter, “Sheng”.
Claim 2. GV fails to explicitly teach selecting, by the processor and based on the particular document category, a second machine-learning (ML) model from a plurality of ML models. Sheng, in the field of document classification and extraction using machine learning, teaches The method of claim 1, wherein the ML model is a first ML model, and determining the data field comprises: selecting, by the processor and based on the particular document category, a second machine-learning (ML) model from a plurality of ML models, [0010] the one or more extraction machine learning models may be selected from a plurality of extraction machine learning models based on the determined document category
each model of the plurality of ML models being trained using text data associated with a respective document category; [0005] one or more so-called “category” machine learning models may be trained, e.g., using a corpus of labeled structured documents, to classify structured documents into one or more of a plurality of document categories (or “verticals”).
[0032] Machine learning application engine 252 may be configured to apply, as labeled training data, features of each feature vector as input to one or more category machine learning models 254. Based on this training data, each category machine learning model 254 may be trained to classify structured documents into one or more of a plurality of document categories (or “verticals”).
[0037] each category machine learning model 254 may be trained as described above to determine whether a structured document 300 should be classified into a particular category
inputting, by the processor, the text segment to the second ML model, the second ML model being trained to receive text data as input, [0037] A cluster 132.sub.x of structured documents 300 associated with (e.g., used to generate, sharing fixed content with, etc.) a given data extraction template 134.sub.x are provided as input to feature extraction engine 250. Feature extraction engine 250 may extract various features <f.sub.1, f.sub.2, . . . , f.sub.n> from each structured document 300 of cluster 132.sub.x and may provide those features to machine learning application engine 252.
and output identification of a corresponding data field; [0033] Machine learning application engine 252 may be further configured to apply, as labeled training data, features of each feature vector as input to one or more extraction machine learning models 256. Based on this training data, each extraction machine learning model 256 may be trained to provide (e.g., classify) one or more locations of one or more transient fields in structured documents.
and receiving, by the processor, as output from the second ML model, the data field corresponding to the text segment. [0033] each transient field location (e.g., XPath) in a structured document that contains transient data of interest may be labeled, annotated, or otherwise indicated.
Thus, before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of GV with the teachings of Sheng [0001-0002] to alleviate the manual burden of classifying a document.
Claim 19. GV fails to explicitly teach identify, based on the text data, an existing digital record of the claim processing system. Sheng, in the field of document classification and extraction using machine learning, teaches The non-transitory computer-readable medium of claim 17, wherein the instructions further cause the processor to: identify, based on the text data, an existing digital record of the claim processing system; [0008] Once associations between the data extraction template and one or more categories/transient field locations are stored (e.g., as annotations of the template, or separately in a database) (understood to be existing digital record), the data extraction template may be used to quickly extract pertinent information from subsequent structured documents… at least one data point may be extracted from the subsequent structured document based on a previously-determined association between the data extraction template and one or more transient field locations(understood to be existing digital record)
and determine, as a value of the data field, a value represented by the text segment. [0003] As used herein, a “transient” field is a location within a cluster of structured documents that does not contain fixed content, instead containing content that is typically directed to a particular user (e.g., price, departure data, invoice number, etc.) (understood to be the value). The classified category(s) and transient field location(s) may then be associated with the data extraction template, e.g., as annotations of the template, data stored within the template, etc.
[0008] at least one data point may be extracted from the subsequent structured document based on a previously-determined association between the data extraction template and one or more transient field locations(understood to be existing digital record). For example, the data extraction template may include one or more annotations indicating that an “event location” is located at a first XPath, an “event start time” is located at a second XPath, and an “event title” is located at a third XPath. These annotations may be used to extract the data located at the designated XPaths.
Thus, before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of GV with the teachings of Sheng [0001-0002] to alleviate the manual burden of classifying a document.
Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Document Classification and Information Extraction framework for Insurance Application to GV et al., hereinafter, “GV” and in view of US 2024/0370827 A1 to Zionpour et al., hereinafter, “Zionpour”.
Claim 8. GV fails to explicitly teach selecting, by the processor and based on the particular document category, a second machine-learning (ML) model from a plurality of ML models. Zionpour, in the field of text extraction from a document, teaches The method of claim 1, further comprising: determining, by the processor and based at least in part on the particular document category, an action to be performed with the electronic document, wherein the action comprises at least one of: associating the electronic document with a priority level, setting permissions of the electronic document to allow the electronic document to be externally viewable, routing the electronic document to a destination for further processing, updating a dashboard to indicate receipt of the electronic document, or linking the electronic document to an digital record of the claim processing system. [0153] Restricting access to entities possessing permission for access to a particular block may include performing a lookup of authorized entities in a repository with respect to the particular block and enabling the authorized entities to view and/or interact with the information contained in the particular block. In response to determining that an entity lacks authorization to access the particular block, the system may omit display of information in the particular block from the unauthorized entity or otherwise prevent the unauthorized entity from interacting with the information in the particular block. For example, the particular block may be associated with (e.g., may include metadata relating to) at least one account identifier, device identifier, network identifier, group identifier, user identifier, or other information delineating at least one criterion, which, when satisfied, causes access to information within the particular block (e.g., an instance of the electronic non-word processing application).
Thus, before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of GV with the teachings of Zionpour [0002] to optimize the time and resources on collaborative projects.
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Document Classification and Information Extraction framework for Insurance Applications to GV et al., hereinafter, “GV” and in view of US 2022/0237373 A1 to Singh Bawa et al., hereinafter, “Singh Bawa” in view of US 2009/0116757 A1 to Neogi et al., hereinafter, “Neogi”.
Claim 9. GV fails to explicitly teach a set of category-specific ML models. Singh Bawa, in the field of document classification and extraction using machine learning, The method of claim 1, wherein the ML model: comprises a set of category-specific ML models, each category-specific model of the set of category-specific ML models being configured to output: [0007] first feature data extracted from the input data to a first set of machine learning (ML) models to select a document category corresponding to the document from a plurality of predefined document categories. The first set of ML models are configured to categorize documents into the plurality of predefined document categories based on input feature data.
and an associated confidence score, provides, as input, to each category-specific model of the set of category-specific ML models, [0029] the first ML models 130 may be configured to output probability scores indicating the estimated probability that an input document corresponds to each of the predefined document categories,
the text data, and [0030] The annotator 132 is configured to perform category-specific annotation (e.g., to generate annotation data) based on input documents… annotation data that includes the names (e.g., identifiers)
generates, as output, the respective document category determined by a category-specific ML model, the category-specific ML model associated with a highest confidence score of the set of category-specific ML models. [0029] the first ML models 130 may be configured to output probability scores indicating the estimated probability that an input document corresponds to each of the predefined document categories, and the input document may be assigned to the predefined document category associated with the greatest probability score.
Thus, before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of GV with the teachings of Singh Bawa [0002] to reduce the time and expense of classifying a document.
GV fails to explicitly teach a binary determination of whether an input document corresponds to a respective document category. Neogi, n the same field of document classification, teaches a binary determination of whether an input document corresponds to a respective document category, [0151] If the system is not successful in finding a class match within a defined threshold, the document is marked as unclassified by the success evaluation module defined above. Examiner interprets not successful/success as binary determination.
Thus, before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of GV with the teachings of Neogi [0008-0012] to reduce the time and expense of classifying a document.
Allowable Subject Matter
Claims 3-7, 13 and 16 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The innovation that makes claims 3 and 16 is “determining, by the processor and from the text data, a second text segment indicative of a particular digital record of the claim processing system; determining, by the processor, that a first data field of the particular digital record matching the data field corresponding to the first text segment is empty;
and inserting, by the processor, the first text segment as a value of the first data field in the particular digital record.
Likewise claim 4 would be allowable because it depends on claim 3.
The innovation that makes claim 5 is “and based on determining that the second electronic document is of the image type, determining, by the processor and using a third ML model configured to output a scene class corresponding to an input image, a scene class corresponding to the second electronic document.”
Likewise claims 6 and 7 would be allowable because it depends on claim 5.
The innovation that makes claim 13 allowable is “based on determining that the particular document category received as the output of the ML model does not match the actual document category, augmenting the training data-set to include the text data and the actual document category; and training the ML model with the augmented training data-set.”
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. US 2021/0157861 A1 to Katzman et al. and US 2017/0032189 A1 to Zhang et al.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DELOMIA L GILLIARD whose telephone number is (571)272-1681. The examiner can normally be reached 8am-5pm.
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, John Villecco can be reached at (571) 272-7319. 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.
/DELOMIA L GILLIARD/Primary Examiner, Art Unit 2661
Prosecution Insights