Knowledge-Enhanced Chain-of-Thought Prompting for Structured Reasoning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Natural language processing systems face challenges in structured reasoning tasks, with pre-trained language models (PLMs) lacking knowledge coverage and knowledge graphs providing inconsistent results, while chain-of-thought prompting improves accuracy but lacks structured knowledge.
Innovation Solution
A method that generates natural language logic paths between candidate answers and questions using a knowledge graph, forming a prompt with these paths and PLM inputs to enhance reasoning capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If pre-trained language models (PLMs) are used for question answering, then broad knowledge coverage is achieved, but structured reasoning capability deteriorates
Solution Approach 1:
The patent combines PLMs and knowledge graphs into a unified system where the PLM processes natural language input and the KG provides structured knowledge. The system merges the broad knowledge coverage of PLMs with the structured reasoning capability of KGs by integrating their outputs through a scoring mechanism that evaluates both semantic similarity and logical path consistency.
Solution Approach 2:
The patent introduces logic paths as an intermediary element that bridges the PLM and KG components. These logic paths represent structured reasoning trajectories through the knowledge graph, serving as a mediator that translates KG structure into form understandable by the PLM and enabling explainable predictions while maintaining structured reasoning capability.
2Reliability
If knowledge graphs are used for structured reasoning, then explainable predictions are enabled, but knowledge coverage deteriorates
Solution Approach 1:
The system merges the explainable prediction capability of knowledge graphs with the broad knowledge coverage of pre-trained language models. The PLM component provides access to extensive world knowledge while the KG component provides explainable reasoning paths, and their combination achieves both broad coverage and explainability simultaneously.
3Reliability
If chain-of-thought prompting is used to unlock reasoning capabilities, then prediction performance is improved, but structured knowledge integration deteriorates
Solution Approach 1:
The patent uses logic paths as an intermediary that enables structured knowledge integration while maintaining improved prediction performance. The logic paths provide a structured representation of reasoning trajectories that the PLM can process, allowing the system to integrate structured knowledge from the KG without sacrificing the prediction performance improvements gained from chain-of-thought approaches.
Data Source
AI summary
Methods and systems for processing a natural language input query that includes a question and a respective set of candidate answers for the question, including generating, based on the input query and a knowledge graph, natural language logic paths between at least some of the candidate answers and the question; forming a natural language prompt based on both the input query and the logic paths; and obtaining a response from a pretrained natural language processing model based on the natural language promp.