Knowledge Graph Construction Using Prior Knowledge Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional knowledge graph construction methods rely heavily on manual annotation, leading to high costs, low accuracy, and incompleteness, and large language models struggle with unreliable data and poor domain-specific understanding.
Innovation Solution
An automatic knowledge graph construction method using prior knowledge and knowledge connection, involving data retrieval, injection of prompt templates into large language model agents, and multi-round feedback to generate entity-relation-entity triples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation and manual collation of expert knowledge are used, then knowledge graph accuracy is improved, but construction cost and time increase significantly
Solution Approach 1:
The patent introduces an intermediary verification mechanism where extracted knowledge triples are validated against pre-stored prior knowledge before being integrated into the knowledge graph. This intermediary step filters unreliable data while maintaining high accuracy, avoiding the need for time-consuming manual verification of each triple.
Solution Approach 2:
The patent performs preliminary action by storing and organizing prior knowledge domains and entities before the actual knowledge graph construction process. This pre-prepared knowledge base enables rapid verification and filtering during extraction, significantly reducing construction time while maintaining accuracy standards.
2Productivity
If large language models are used for automatic knowledge graph construction, then construction efficiency is improved, but data reliability and domain knowledge accuracy deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the large language model's extracted triples are continuously validated against the stored prior knowledge. Inconsistent or unreliable extractions are identified and corrected through this feedback loop, maintaining data reliability while preserving the efficiency benefits of automatic extraction.
Solution Approach 2:
The patent applies preliminary anti-action by pre-storing accurate domain knowledge and using it to counteract potential errors in large language model extractions before they contaminate the knowledge graph. This preventive approach maintains reliability without sacrificing the speed of automatic construction.
3Ease of operation
If general large language models are used, then ease of operation is improved, but domain-specific knowledge completeness and precision worsen
Solution Approach 1:
The patent applies local quality by storing and utilizing domain-specific prior knowledge tailored to each knowledge graph construction task. This localized knowledge enhancement allows the use of general large language models while achieving domain-specific precision through the targeted application of relevant prior knowledge during the verification process.
Data Source
AI summary
An automatic knowledge graph construction method based on prior knowledge and knowledge connection is disclosed. The method includes obtaining prompt data by storing relevant topic information for constructing a knowledge graph as character strings; retrieving and saving article paragraphs from external data source based on the prompt data; respectively injecting prompt templates for four large language model agents of annotation, reasoning, cognition, and association; obtaining prior knowledge by inputting the injected prompt templates, article paragraphs, and specific task requirements into the agents; obtaining effective data related to the knowledge graph topic by inputting the article paragraphs, prior knowledge, and a pre-defined contrasting prompt text into a knowledge-connecting large language model; and configuring the effective data related to the knowledge graph topic and pre-defined input prompts as an input layer of a large language model automatic agent framework, obtaining entity-relation-entity triples, and completing the construction of the knowledge graph.