Self-VEQA Agent Self-Verification Enhanced Question Answering Agent
Abstract: The task of Knowledge Graph Question Answering (KGQA) involves using information stored in a knowledge graph (KG) to answer questions by identifying the relation path between the subject entity and the answer. Traditional KGQA methods require extensive training data and are time-consuming. Recent advancements in Large Language Models (LLMs) have shown potential in various tasks. However, methods leveraging LLMs for KGQA face challenges such as inference errors and excessive reliance on prompt design. To address these issues, we propose the Self-VEQA Agent, which utilizes two agents: a QA Agent for initial answers based on KG and a Verification Agent to iteratively refine these answers, improving accuracy over time. Additionally, our model features a memory mechanism that enables dynamic evolution. As the Self-VEQA Agent performs tasks and accumulates experience, the overall performance improves over time. Evaluated on two KGQA benchmarks, Self-VEQA Agent outperforms most traditional and LLM-based methods, demonstrating its effectiveness.
Paper Type: Long
Research Area: Question Answering
Research Area Keywords: logical reasoning; knowledge base QA; multihop QA; interpretability; reasoning; few-shot QA; open-domain QA;
Contribution Types: Approaches to low-resource settings
Languages Studied: English
Submission Number: 575
Loading