Knowledge Graph Prompt for AI Reasoning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Pretrained language models face difficulties in reasoning-related tasks when required knowledge is absent from their training corpus or when test instances are not in question-answering format, and they struggle to provide explanations for their predictions.
Innovation Solution
A method that labels text data from inquiries, links it to a knowledge graph entity, retrieves relevant reasoning paths, removes irrelevant paths using a language model, and employs the remaining paths to provide answer predictions, enabling decision support in domains like healthcare without requiring task-specific supervision or answer candidates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If pretrained language models are used for reasoning tasks, then they can process general text data, but they fail when required knowledge is absent from training corpus or test instances are not in question-answering format
Solution Approach 1:
The patent introduces a knowledge graph as an intermediary component between the pretrained language model and the reasoning task. The knowledge graph stores structured domain knowledge and reasoning paths, allowing the system to retrieve relevant information even when the knowledge is not present in the training corpus of the language model. This mediator enables the system to handle out-of-distribution questions by leveraging external knowledge sources.
Solution Approach 2:
The patent segments the reasoning system into distinct functional components: a pretrained language model for natural language processing, a knowledge graph for structured knowledge storage, and a retrieval mechanism for accessing reasoning paths. This segmentation allows each component to specialize in specific tasks, with the knowledge graph handling factual reasoning and the language model handling natural language interaction, thereby improving overall adaptability and reliability.
2Productivity
If pretrained language models are used for reasoning tasks, then they can generate predictions, but they base predictions on implicitly encoded knowledge that is incapable of handling structured reasoning and does not offer explanations
Solution Approach 1:
The knowledge graph serves as an intermediary that stores not only factual knowledge but also structured reasoning paths and relationships. When generating predictions, the system retrieves relevant reasoning paths from the knowledge graph, providing explicit explanations for the predictions. This mediator bridges the gap between fast prediction generation and provideable explanations by storing reasoning trajectories that can be retrieved and presented to the user.
Solution Approach 2:
The system performs preliminary retrieval of relevant reasoning paths from the knowledge graph before generating the final prediction. This preliminary action ensures that the language model has access to structured reasoning information and relevant context, enabling it to generate predictions that are both fast and explainable. The reasoning paths are pre-computed and stored, allowing quick retrieval during prediction time.
Data Source
AI summary
A computer-implemented method for optimized decision making that includes labeling text data extracted from an inquiry, and linking labeled text to a knowledge graph entity. The method may further include retrieving from the knowledge graph reasoning paths; and removing irrelevant knowledge graph reasoning paths using a language model trained artificial intelligence consistent with the labeling of the text data. The method may further include employing remaining relevant graph reasoning paths to provide an answer prediction.