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 .
Claims 1-20 are pending. Claims 1 and 11 are independent.
Apparent priority 10/31/2024.
This action is Non-Final.
Claim Rejections - 35 USC § 102
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)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-9 and 11-19 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Larson (US PG Pub 20250328565).
As per claims 1 and 11, Larson discloses:
A computing system comprising: a processor (Larson; Fig. 24, item 2404; p. 0260 - the computer 2402 (computing system) including a processing unit 2404 (processor)); and a non-transitory computer-readable storage device storing computer-executable instructions, the instructions when executed by the processor cause the processor to perform operations (Larson; Fig. 24; p. 0263 - The drives and their associated computer-readable storage media provide nonvolatile storage of data, data structures, computer-executable instructions, and so forth… any such storage media can contain computer-executable instructions for performing the methods described herein) comprising: receiving a query from a user device (Larson; Fig. 3, item 304; p. 0106 - …plain text question 304…; see also p. 0162); embedding the received query to a vector space (Larson; Fig. 11; p. 0162 - …the search component 320 can execute the encoder portion 308 on the plain text question 304, thereby yielding an embedding (e.g., a latent vector representation) for the plain text question 304…; see also p. 0046-0047) comprising a plurality of embedded documents (Larson; p. 0052-0055 - …the document-graph repository can contain a plurality of document-graphs that respectively correspond to the plurality of technical documents… each context-tagged text block (e.g., each concatenation of downstream text block and upstream contextual information) in the document-graph repository can have its own respective embedding (e.g., its own respective latent vector)); extracting one or more entities from the received query (Larson; p. 0163 - the search component 320 can identify one or more keywords within the plain text question 304 (e.g., via execution of the named entity recognition neural network 812 or any other named entity extraction tool); see also p. 0074 & 0202) based on a pre-defined configuration of entities (Larson p. 0061 - …by leveraging the named entity recognition neural network, the search component can identify the explicitly-recited scientific instrument identifiers for each given technical document… pre-defining configuration of entities using instrument identifiers; see also p. 0142 - the named entity recognition neural network 812 can be configured to determine what specific scientific instrument identifiers are recited within an inputted textual document and, in some cases, where those specific scientific instrument identifiers are located in the inputted textual document. So, the search component 320 can electronically execute the named entity recognition neural network 812 on the technical document 600, and such execution can yield one or more instrument identifiers 814); generating a hybrid query based on the extracted entities and the received query (Larson; p. 0163 - the search component 320 can identify one or more keywords within the plain text question 304 (e.g., via execution of the named entity recognition neural network 812 or any other named entity extraction tool), and the search component 320 can search for context-tagged text blocks in the document-graph repository 402 that recite or otherwise contain those same one or more keywords; see also p. 0074 - if the plain text question contains one or more scientific instrument identifiers (e.g., which can be determined via execution of the named entity recognition neural network on the plain text question), then the search component can augment the unified prompt so as to emphasize those one or more scientific instrument identifiers; see also p. 0202); pre-filtering the vector space based on the extracted entities (Larson; p. 0037 - tagging each text block with its corresponding contextual information can help to eliminate the occurrence of no-identifier text blocks (pre-filtering). Accordingly, context-tagging as described herein can help to avoid situations in which RAG-AI mistakenly or erroneously synthesizes answers for certain scientific instruments using text blocks that are not applicable or pertinent to those certain scientific instruments. In this way, context-tagging via document-graphs as described herein can help to increase or boost answer accuracy of RAG-AI); identifying one or more documents relevant to the hybrid query from the pre-filtered vector space (Larson; Fig. 11; p. 0161-0167 - …the search component 320 can electronically perform a search through the document-graph repository 402, and such search can uncover or otherwise identify a plurality of potentially-relevant context-tagged text blocks 1102…); generating an input based on the user query and information parsed from the one or more identified documents (Larson; Fig. 4, Fig. 17 & Fig, 19, item 406; p. 0118 - …the search component 320 can aggregate the plain text question 304 and the plurality of relevant context-tagged text blocks 404 together, thereby forming a unified prompt 406; see also p. 0201 & p. 0206-0211; see also p. 0040); analyzing the input with a large language model (LLM) (Larson; Fig. 19, item 306; p. 0207-0208 - the search component 320 can feed the unified prompt 406 to the LLM 306, the unified prompt 406 can complete a forward pass through the LLM 306 (e.g., through the encoder portion 308 and the synthesizer portion 310); see also p. 0081; see also p. 0040); receiving a response to the query from the LLM (Larson; Fig. 19, item 1802; p. 0207-0208 - …the LLM 306 can compute or calculate the plain text answer 1802 based on whatever activation maps or feature maps were internally generated by the LLM 306 during such forward pass…; see also p. 0081; see also p. 0048); and transmitting the response to the user device (Larson; Fig. 19, item 1802; p. 0207-0208 - …the answer component 322 can electronically transmit the plain text answer 1802 to any suitable computing device…; see also p. 0081).
As per clams 2 and 12, Larson discloses: The computing system and method of claims 1 and 11, wherein identifying the one or more documents relevant to the hybrid query comprises performing a similarity analysis on the hybrid user query and the plurality of embedded documents (Larson; p. 0162 - the search component 320 can search for context-tagged text blocks in the document-graph repository 402 whose embeddings are within any suitable threshold margin of similarity (e.g., in terms of Euclidean distance or cosine similarity) of the embedding of the plain text question 304; see also p. 0169 & p. 0039).
As per claims 3 and 13, Larson discloses: The computing system and method of claims 2 and 12, wherein performing the similarity analysis comprises performing a cosine similarity ranking of embedded documents within the plurality of embedded documents (Larson; p. 0162 - the search component 320 can search for context-tagged text blocks in the document-graph repository 402 whose embeddings are within any suitable threshold margin of similarity (e.g., in terms of Euclidean distance or cosine similarity) of the embedding of the plain text question 304; see also p. 0169 & p. 0039 - …To reconcile or compare these differently-discovered context-tagged text blocks, a re-ranker can be implemented to assign to each discovered or found context-tagged text block a relevance score showing how relevant or irrelevant a respective context-tagged text block is to the given natural language question…).
As per claims 4 and 14, Larson discloses: The computing system and method of claims 2 and 12, wherein performing the similarity analysis comprises identifying and ranking a predefined number of relevant embedded documents based on a relevance to the query (Larson; p. 0162 - the search component 320 can search for context-tagged text blocks in the document-graph repository 402 whose embeddings are within any suitable threshold margin of similarity (e.g., in terms of Euclidean distance or cosine similarity) of the embedding of the plain text question 304; see also p. 0169 & p. 0039 - …To reconcile or compare these differently-discovered context-tagged text blocks, a re-ranker can be implemented to assign to each discovered or found context-tagged text block a relevance score showing how relevant or irrelevant a respective context-tagged text block is to the given natural language question…).
As per clams 5 and 15, Larson discloses: The computing system and method of claims 1 and 11, wherein the operations further comprise: receiving a list of pre-defined entities from the user device; and adding the list of pre-defined entities to the pre-defined configuration (Larson p. 0061 - …by leveraging the named entity recognition neural network, the search component can identify the explicitly-recited scientific instrument identifiers for each given technical document… pre-defining configuration of entities using instrument identifiers).
As per clams 6 and 16, Larson discloses: The computing system and method of claims 1 and 11, wherein the pre-defined configuration is persisted as metadata within the vector space (Larson; p. 0078 - or a first document-graph of the plurality of document-graphs that corresponds to a first technical document of the plurality of technical documents, leaf nodes of the first document-graph can represent respective text blocks written in the first technical document, and non-leaf nodes of the first document-graph can respectively represent a document title, one or more section headings, and one or more scientific instrument identifiers written in the first technical document (metadata) and beneath which the respective text blocks are nested).
As per clams 7 and 17, Larson discloses: The computing system and method of claims 1 and 11, wherein the operations further comprise: accessing the plurality of embedded documents; extracting one or more keywords from the plurality of embedded documents; and persisting the one or more extracted keywords as metadata within the vector space (Larson; p. 0059 - the named entity recognition neural network can be configured to extract specific types of named entities from any given inputted document. In particular, the named entity recognition neural network can be configured to extract scientific instrument identifiers from any given inputted document. Accordingly, for each of the plurality of technical documents, the search component can execute the named entity recognition neural network on the technical document, and such execution can yield one or more specific scientific instrument identifiers that are written somewhere within that technical document; see also p. 0078).
As per clams 8 and 18, Larson discloses: The computing system and method of claims 7 and 17, wherein extracting the one or more keywords from the plurality of embedded documents comprises extracting the one or more keywords using the LLM (Larson; p. 0054 - the search component can generate such embedding by leveraging the encoder portion of the LLM. In other words, for any given context-tagged text block, the search component can execute the encoder portion of the LLM on that given context-tagged text, and such execution can yield an embedding for that given context-tagged text block).
As per clams 9 and 19, Larson discloses: The computing system and method of claims 8 and 18, wherein extracting the one or more keywords using the LLM comprises extracting the one or more keywords in a zero-shot fashion (Larson; p. 0054 - the search component can generate such embedding by leveraging the encoder portion of the LLM. In other words, for any given context-tagged text block, the search component can execute the encoder portion of the LLM on that given context-tagged text, and such execution can yield an embedding for that given context-tagged text block).
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 10 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Larson in view of He (US PG Pub 20240362286).
As per clams 10 and 20, Larson discloses: The computing system and method of claims 1 and 11, wherein generating the hybrid query comprises generating a query comprising a lexical clause, a semantic clause, and a keyword clause (Larson; p. 0163 - the search component 320 can identify one or more keywords within the plain text question 304 (e.g., via execution of the named entity recognition neural network 812 or any other named entity extraction tool), and the search component 320 can search for context-tagged text blocks in the document-graph repository 402 that recite or otherwise contain those same one or more keywords; see also p. 0074 - if the plain text question contains one or more scientific instrument identifiers (e.g., which can be determined via execution of the named entity recognition neural network on the plain text question), then the search component can augment the unified prompt so as to emphasize those one or more scientific instrument identifiers; see also p. 0202). Larson, however, fails to disclose wherein generating the hybrid query comprises generating a query comprising a lexical clause, a semantic clause. He does teach wherein generating the hybrid query comprises generating a query comprising a lexical clause, a semantic clause (He; p. 0036 - an electronic document management system may implement a set of improved search tools and algorithms to perform lexical searching, semantic searching, or a combination of both). Therefore, it would have been obvious to one of ordinary skill in the art to modify the system and method of Larson to include wherein generating the hybrid query comprises generating a query comprising a lexical clause and a semantic clause, as taught by He, because lexical searching can be useful for identifying instances of a specific keyword or phrase in a large dataset, to extract relevant information from unstructured text, or to monitor online conversations for particular topics or keywords (He; p. 0029), while semantic searching can be very helpful in quickly locating the relevant information within an electronic document, such as an electronic agreement. It saves time compared to manually going through the entire document and it can be especially useful in cases where the document is very long or complex (He; p. 0036).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. The prior art made of record and not relied upon includes: Brown (US PG Pub 20250363345) discloses a large language model (LLM) may be used to classify an input into one of a plurality of categories. However, given the machine-learning operation of the LLM, the output of the LLM does not represent a definitive statement, but is based on probability computations of the machine learning model. Therefore, the classification performed by the LLM might not be correct. Classification into the wrong category by the LLM results in downstream technical problems. In some implementations, when an LLM generates a response that classifies an input, one or more probability values associated with a token that forms the basis of the response may be used to determine a confidence value. The confidence value is indicative of confidence in the classification performed by the LLM. An action may be taken based on the confidence value (Brown; Abstract).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Rodrigo A Chavez whose telephone number is (571)270-0139. The examiner can normally be reached Monday - Friday 9-6 ET.
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, Richemond Dorvil can be reached at 5712727602. 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.
/RODRIGO A CHAVEZ/Examiner, Art Unit 2658
/RICHEMOND DORVIL/Supervisory Patent Examiner, Art Unit 2658
Prosecution Insights