Enhanced Language Model for Complex Entity Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current knowledge extraction techniques from unstructured data are limited to understanding simple grammatical or semantic constructs, unable to extract complex entities and relationships, as they only yield simple triples that fail to uncover hidden relationships in the data.
Innovation Solution
The method involves parsing unstructured data using an existing parser to generate a parse tree, identifying tokens, and enhancing the language model by re-inserting these tokens to create a knowledge graph, which collapses branches into more complex relationships, allowing for the extraction of complex entities and their inter-relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple triple extraction methods are used, then the extraction process is simple and fast, but the quality and depth of extracted knowledge is limited
Solution Approach 1:
The patent segments the extraction process into multiple stages: initial triple extraction, token identification, language model enhancement, and re-parsing. This multi-stage segmentation allows the system to progressively extract increasingly complex relationships while maintaining manageable processing at each stage, resolving the contradiction between extraction depth and process complexity
Solution Approach 2:
The patent performs preliminary actions by first extracting simple triples and identifying tokens before enhancing the language model. This preliminary extraction of basic entities and relationships prepares the foundation for subsequent complex relationship extraction, allowing the system to build up extraction capability incrementally rather than attempting complex extraction directly
2Loss of information
If complex entities and relationships are extracted, then the quality of machine usable information improves, but the processing time and computational resources increase
Solution Approach 1:
The system performs preliminary extraction of tokens and simple relationships before enhancing the language model and re-parsing. This preliminary action prepares the data structure in advance, reducing the computational burden during the more intensive complex relationship extraction phase and thereby reducing overall processing time
Solution Approach 2:
The patent creates copies of the language model enhanced with extracted tokens, allowing multiple re-parsing operations to be performed on enhanced models without repeatedly modifying the original. This copying approach enables efficient iterative extraction of complex relationships while minimizing redundant processing
Data Source
AI summary
To extract relationships between complex entities from unstructured data, a parser parses, using an existing language model, the unstructured data to generate a parse tree. From the parse tree, a set of tokens is created. A token in the set of tokens includes a set of words found in the unstructured data. The set of tokens is inserted in the existing language model to form an enhanced language model. The unstructured data is re-parsed using the enhanced language model to create a knowledge graph. From the knowledge graph, a relationship between a subset of the set of tokens is extracted.