Knowledge Graph Updating Through Context-Aware Language Model Prompts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of efficiently updating knowledge graphs for facilities or infrastructures using language models is hindered by the need for manual intervention, high human and temporal costs, and the complexity of maintaining and transferring domain-specific knowledge, especially when documents are incomplete or fragmented, and the language model lacks specialized training.
Innovation Solution
A system that utilizes a document information accepting section, knowledge graph information accepting section, prompt constructing section, prompt presenting section, response information accepting section, and update query constructing section to automate the process of updating knowledge graphs by constructing prompts based on both document and knowledge graph information, ensuring context-based updates even with incomplete documents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual updating of knowledge graph is performed based on document contents, then knowledge graph can be updated with domain knowledge, but human cost and temporal cost increase
Solution Approach 1:
The patent replaces manual mechanical updating processes with an automated system that uses language models to construct prompts and generate update queries. The system automatically processes document information, constructs appropriate prompts, and generates knowledge graph update queries without requiring manual intervention, thereby reducing temporal cost while maintaining update accuracy
Solution Approach 2:
The system enables self-service updating where the knowledge graph system automatically processes document information and generates its own update queries. The language model constructs prompts and generates update queries autonomously based on the document contents, eliminating the need for human operators to manually create update queries
2Reliability
If manual updating of knowledge graph is performed, then domain knowledge can be incorporated, but human cost increases
Solution Approach 1:
The patent replaces manual human operations with an automated system that uses language models to process document information, construct prompts, and generate update queries. This substitution maintains domain knowledge accuracy while achieving high automation level
Solution Approach 2:
The language model acts as an intermediary between the document information and the knowledge graph update process. It constructs prompts that bridge the gap between raw document contents and structured knowledge graph updates, enabling automated processing while maintaining domain knowledge accuracy
3Extent of automation
If language model is used for knowledge graph update, then automation level increases, but system complexity increases
Solution Approach 1:
The patent segments the knowledge graph update process into distinct functional components: document information processing, prompt construction, language model inference, and update query generation. This segmentation allows each component to be optimized independently and simplifies the overall system architecture despite using language models
Solution Approach 2:
The language model serves as an intermediary component that handles the complex task of transforming document information into update queries. By isolating the language model as a separate intermediary layer, the system manages complexity through modular architecture rather than integrating all functions into a single complex system
4Reliability
If fine tuning technique is used for language model, then domain knowledge can be learned, but retraining cost increases when training data is enormous or frequently modified
Solution Approach 1:
The system performs preliminary action by using a pre-trained language model that already contains general knowledge, rather than requiring retraining for each domain update. The pre-trained model serves as a foundation that can be efficiently fine-tuned or prompted for specific domain tasks without costly retraining
Solution Approach 2:
Instead of changing the fundamental model parameters through retraining, the system uses parameter changes in the form of prompts and fine-tuning adjustments. This allows the language model to adapt to domain knowledge through prompt engineering and minor parameter adjustments rather than costly retraining of the entire model
Data Source
AI summary
An object is to realize updating based on the context of a knowledge graph when a document triggers updating of the knowledge graph or realize updating even if the document has been subjected to omission or the like. A document information accepting section accepts document information which is information included in a document that triggers updating of a knowledge graph. A knowledge graph information accepting section accepts knowledge graph information from a knowledge graph system. A prompt constructing section constructs a prompt on the basis of the document information and the knowledge graph information. A response information accepting section accepts response information representing a response from the language model system to the prompt. An update query constructing section constructs the update query on the basis of the response information. An update query presenting section presents the update query to the knowledge graph system.