# US 12,217,007 B2 · Providing a semantic encoding and language neural network > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,217,007 B2 - Document ID: US12217007B2 - Status: Granted - Kind: B2 - Application number: 17/811,763 - Filed: 2022-07-11 - Published: 2024-01-11 - Granted: 2025-02-04 - Earliest claimed: 2022-07-11 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Thanh Lam Hoang; Dzung Phan; Gabriele Picco; Lam Nguyen; Marco Luca Sbodio; Vanessa Lopez Garcia - Web page: https://patentgrep.com/patent/US12217007B2 ## CPC - G06F 40/30: Semantic analysis - G06F 40/126: Character encoding - G06F 40/205: Parsing - G06F 40/279: Recognition of textual entities - G06F 40/40: Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30) - G06N 3/044: Recurrent networks, e.g. Hopfield networks - G06N 3/045: Combinations of networks - G06N 3/08: Learning methods - G06N 3/088: Non-supervised learning, e.g. competitive learning - G06N 5/022: Knowledge engineering; Knowledge acquisition - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Embodiments are provided for unsupervised learning of domain specific knowledge graph from textual data and language generation from knowledge graph via reinforcement learning in a computing system by a processor. Unstructured data is automatically parsed into one or more knowledge graphs based on the unstructured data and a list of candidate relations using a first machine learning model. Text data is generated from the one or more knowledge graphs using a second machine learning model. ## Claims ### Claim 1 A method for providing semantic encoding and language generation in a computing system by a processor, comprising: automatically parsing unstructured data into one or more knowledge graphs based on the unstructured data and a list of candidate relations using a first machine learning model; encoding, using the first machine learning model, the unstructured data into a distribution of a plurality of triples based on the one or more knowledge graphs, wherein the encoding further comprises predicted probabilities of relations between entities in the unstructured data; sampling, using a second machine learning model, a set of the plurality of triples from the unstructured data of the one or more knowledge graphs; generating text data from the set of the plurality of triples using the second machine learning model; computing a penalty score for the set of the plurality of triples based on a degree of difference between the unstructured data and the generated text data; and adjusting at least one predicted probability from the first machine learning model based on the determined penalty score. ### Claim 2 (depends on 1) The method of claim 1, further including training the first machine learning model and the second machine learning model using the unstructured data and the list of candidate relations via unsupervised machine learning, wherein the first machine learning model is a semantic encoder and the second machine learning model is a semantic decoder. ### Claim 3 (depends on 1) The method of claim 1, further including using the first machine learning model to: identify the entities in the unstructured data. ### Claim 4 (depends on 1) The method of claim 1, further including using the second machine learning model to: decode the set of the plurality of triples into the text data, wherein a triple includes a subject, object, and predicate in the unstructured data, wherein the subject and object are an entity and a predicate is a relation. ### Claim 5 (depends on 1) The method of claim 1, further including sampling the set of the plurality of triples from the unstructured data of the one or more knowledge graphs for training a plurality of machine learning models via unsupervised machine learning. ### Claim 6 (depends on 1) The method of claim 1, further including: identifying one or more candidate entities in the unstructured data; and using the one or more candidate entities as nodes in the one or more knowledge graphs. ### Claim 7 A system for providing semantic encoding and language generation in a computing environment, comprising: one or more computers with executable instructions that when executed cause the system to: automatically parse unstructured data into one or more knowledge graphs based on the unstructured data and a list of candidate relations using a first machine learning model; encode, using the first machine learning model, the unstructured data into a distribution of a plurality of triples based on the one or more knowledge graphs, wherein the encoding further comprises predicted probabilities of relations between entities in the unstructured data; sample, using a second machine learning model, a set of the plurality of triples from the unstructured data of the one or more knowledge graphs; generate text data from the set of the plurality of triples using the second machine learning model; compute a penalty score for the set of the plurality of triples based on a degree of difference between the unstructured data and the generated text data; and adjust at least one predicted probability from the first machine learning model based on the determined penalty score. ### Claim 8 (depends on 7) The system of claim 7, wherein the executable instructions when executed cause the system to train the first machine learning model and the second machine learning model using the unstructured data and the list of candidate relations via unsupervised machine learning, wherein the first machine learning model is a semantic encoder and the second machine learning model is a semantic decoder. ### Claim 9 (depends on 7) The system of claim 7, wherein the executable instructions when executed cause the system to use the first machine learning model to: identify the entities in the unstructured data. ### Claim 10 (depends on 7) The system of claim 7, wherein the executable instructions when executed cause the system to use the second machine learning model to: decode the set of the plurality of triples into the text data, wherein a triple includes a subject, object, and predicate in the unstructured data, wherein the subject and object are an entity and a predicate is a relation. ### Claim 11 (depends on 7) The system of claim 7, wherein the executable instructions when executed cause the system to sample the set of the plurality of triples from the unstructured data of the one or more knowledge graphs for training a plurality of machine learning models via unsupervised machine learning. ### Claim 12 (depends on 7) The system of claim 7, wherein the executable instructions when executed cause the system to: identify one or more candidate entities in the unstructured data; and use the one or more candidate entities as nodes in the one or more knowledge graphs. ### Claim 13 A computer program product for providing semantic encoding and language generation in a computing environment, the computer program product comprising: one or more tangible computer readable storage media, and program instructions collectively stored on the one or more tangible computer readable storage media, the program instruction comprising: automatically parse unstructured data into one or more knowledge graphs based on the unstructured data and a list of candidate relations using a first machine learning model; encode, using the first machine learning model, the unstructured data into a distribution of a plurality of triples based on the one or more knowledge graphs, wherein the encoding further comprises predicted probabilities of relations between entities in the unstructured data; sample, using a second machine learning model, a set of the plurality of triples from the unstructured data of the one or more knowledge graphs; generate text data from the set of the plurality of triples using the second machine learning model; compute a penalty score for the set of the plurality of triples based on a degree of difference between the unstructured data and the generated text data; and adjust at least one predicted probability from the first machine learning model based on the determined penalty score. ### Claim 14 (depends on 13) The computer program product of claim 13, further including program instructions to train the first machine learning model and the second machine learning model using the unstructured data and the list of candidate relations via unsupervised machine learning, wherein the first machine learning model is a semantic encoder and the second machine learning model is a semantic decoder. ### Claim 15 (depends on 13) The computer program product of claim 13, further including program instructions to use the first machine learning model to: identify the entities in the unstructured data. ### Claim 16 (depends on 13) The computer program product of claim 13, further including program instructions to use the second machine learning model to: decode the set of the plurality of triples into the text data, wherein a triple includes a subject, object, and predicate in the unstructured data, wherein the subject and object are an entity and a predicate is a relation. ### Claim 17 (depends on 13) The computer program product of claim 13, further including program instructions to: identify one or more candidate entities in the unstructured data; and use the one or more candidate entities as nodes in the one or more knowledge graphs.