SymAgent: A Neural-Symbolic Self-Learning Agent Framework for Complex Reasoning over Knowledge Graphs
Topics: AI (Deep Learning), Graph RAG, Knowledge Graph, LLMO, Prompt Engineering, Retrieval Augmented Generation (RAG)
This paper introduces SymAgent, a novel framework that combines Large Language Models (LLMs) with Knowledge Graphs (KGs) for complex reasoning tasks. The framework consists of two main components: an Agent-Planner that extracts symbolic rules from KGs to guide reasoning, and an Agent-Executor that interacts with KGs through predefined actions. The paper also presents a self-learning mechanism that allows the system to improve through autonomous interaction, achieving superior performance even with smaller language models compared to larger counterparts like GPT-4.