Bayesian Network Structure Learning for Concept Graph Completeness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for building concept graphs from open domain texts face challenges due to the low co-occurrence of entity-concept pairs, making it difficult to generate complete graphs, as entities and concepts rarely appear together in sentences.
Innovation Solution
A computer-implemented method using Bayesian network structure learning (BNSL) to discover relationships between entities, relations, and concepts from open domain facts, by obtaining entities associated with a concept, searching for relevant facts, and generating positive and negative data observations to learn a Bayesian network structure, which predicts whether a new entity is an instance of a concept.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If pattern-based methods are used to determine entity-concept relations, then the method is simple and easy to implement, but the completeness of concept graphs is poor due to low co-occurrence of entity-concept pairs
Solution Approach 1:
The patent introduces an intermediary approach by using relation extraction as a mediator between entities and concepts. Instead of directly linking entities to concepts through co-occurrence patterns, the system extracts relations between entities and then uses these relations as intermediaries to infer concept associations. This mediator approach allows indirect inference of entity-concept relationships, significantly improving concept graph completeness while maintaining implementation feasibility through structured relation-based reasoning
2Quantity of substance
If relation extraction is performed to improve concept graph completeness, then more entity-concept pairs can be identified, but the complexity of the system increases
Solution Approach 1:
The patent applies segmentation by dividing the concept graph construction task into distinct modular components: entity extraction, relation extraction, and concept association. Each component operates independently on specific data types and can be implemented using separate algorithms. The relation extraction module processes factual statements to extract subject-predicate-object triplets, while the concept association module uses these relations to infer entity-concept connections. This segmentation reduces overall system complexity by making each module manageable and reusable
3Loss of information
If Bayesian network structure learning is used to discover relationships, then interpretable relationships between entities, relations, and concepts can be learned, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-processing the data to extract and structure relations before performing Bayesian network structure learning. The system first extracts relations from factual statements and organizes them into a structured format, then uses this pre-processed relation data as input for Bayesian network learning. This preliminary organization of data reduces the computational burden during the actual Bayesian network inference, making the process more efficient while preserving the interpretability of learned relationships
Data Source
AI summary
Concept graphs are created as universal taxonomies for text understanding in the open domain knowledge. The nodes in concept graphs include both entities and concepts. The edges are from entities to concepts, showing that an entity is an instance of a concept. Presented herein are embodiments that handle the task of learning interpretable relationships from open domain facts to enrich and refine concept graphs. In one or more embodiments, the Bayesian network structures are learned from open domain facts as the interpretable relationships between relations of facts and concepts of entities. Extensive experiments were conducted on English and Chinese datasets. Compared to the state-of-the-art methods, the learned network structures improve the identification of concepts for entities based on the relations of entities on both English and Chinese datasets.