Graph-to-Text Conversion for Language Model Knowledge Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing language models, such as those based on generative pre-trained transformers, are unable to directly process or utilize knowledge graphs due to their unsupervised training on unlabeled human-readable text, limiting their ability to leverage knowledge graphs for training and answering questions.
Innovation Solution
A computer-implemented method is provided to extract grammatically correct sentences from a directed graph by defining conditions and patterns, allowing the computation of strings from the graph's statements, which are then used as input to the language model, thereby maximizing the usability of the knowledge graph for AI applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If language models are trained on unlabeled human-readable text in an unsupervised manner, then training data availability is improved, but the ability to process and utilize knowledge graphs deteriorates
Solution Approach 1:
The patent introduces an intermediary conversion process that transforms knowledge graph data (directed graphs with nodes and edges) into natural language text format. This mediator enables language models trained on unlabeled text to indirectly process knowledge graph information by converting it into a compatible input format, thus resolving the incompatibility between unsupervised training data requirements and knowledge graph processing needs
2Reliability
If knowledge graphs are constructed and maintained with comprehensive enterprise knowledge, then knowledge quality and completeness are improved, but construction and maintenance cost deteriorates
Solution Approach 1:
The patent enables language models to automatically process and utilize knowledge graph data through self-service mechanisms. By converting knowledge graphs into natural language format that models can independently process, the system reduces the need for manual knowledge curation and maintenance, allowing the enterprise knowledge base to serve itself without requiring additional expert intervention for updates and modifications
3Ease of operation
If language models are used to answer questions based on knowledge graphs, then question-answering capability is improved, but direct processing of knowledge graphs by the model deteriorates
Solution Approach 1:
The patent segments the knowledge graph processing task into two distinct parts: (1) conversion of the knowledge graph into natural language text format, and (2) processing of the converted text by the language model. This segmentation allows the language model to focus on its strength (processing natural language) while a separate conversion component handles the knowledge graph structure, improving overall question-answering capability without requiring the model to directly process complex graph structures
Data Source
AI summary
A method, a system and a computer program for providing data from a directed graph to a language model are provided. The method comprises defining a plurality of conditions and a plurality of patterns, wherein each of the conditions has at least one corresponding pattern. The method further comprises receiving a subset of the directed graph, wherein the subset of the directed graph includes a plurality of statements, wherein each of the statements includes a subject, an object and a predicate relating the subject to the object. The method further comprises for each of the statements in the subset of the directed graph, performing the following: when one of the conditions matches a respective statement and the pattern corresponding to the condition can be applied to the respective statement, computing a string from the respective statement using the pattern.