# LightPROF: A Lightweight Reasoning Framework for Large Language Model on Knowledge Graph

**Authors:** Tu Ao, Yanhua Yu, Yuling Wang, Yang Deng, Zirui Guo, Liang Pang, Pinghui Wang, Tat-Seng Chua, Xiao Zhang, Zhen Cai  
**arXiv:** [2504.03137](https://arxiv.org/abs/2504.03137)  
**Date:** 2025-04-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2504.03137v1/

## Key points

- Authors introduce LightPROF as a lightweight reasoning framework for Large Language Models (LLMs) on Knowledge Graphs (KGs)
- LightPROF follows a "Retrieve-Embed-Reason process" to enhance reasoning capabilities of LLMs
- It retrieves relevant reasoning graph accurately from KG, uses Transformer-based Knowledge Adapter to integrate information, and generates LLM-friendly prompts for final tasks
- Experimental results show LightPROF outperforms existing methods with small-scale LLMs in terms of efficiency and performance

## Summary

, , , , 

In their paper "LightPROF: A Lightweight Reasoning Framework for Large Language Model on Knowledge Graph," authors Tu Ao, Yanhua Yu, Yuling Wang, Yang Deng, Zirui Guo, Liang Pang, Pinghui Wang, Tat-Seng Chua, Xiao Zhang, and Zhen Cai introduce a novel approach to enhancing the reasoning capabilities of Large Language Models (LLMs) through the integration of Knowledge Graphs (KGs). The authors propose LightPROF as a lightweight and efficient Prompt learning-ReasOning Framework for KGQA. This framework follows a "Retrieve-Embed-Reason process" where it first retrieves the relevant reasoning graph from the KG accurately and stably through a retrieval module. It then utilizes a Transformer-based Knowledge Adapter to extract and integrate factual and structural information from the KG before mapping this information to the token embedding space of the LLM. This process generates an LLM-friendly prompt for final reasoning tasks. Notably, LightPROF only requires training of the Knowledge Adapter and is compatible with any open-source LLM model. Experimental results on two public KGQA benchmarks demonstrate that LightPROF outperforms existing methods when used with small-scale LLMs. Furthermore, LightPROF exhibits advantages in terms of input token count and reasoning time efficiency. The paper has been accepted by AAAI 2025 and showcases how leveraging KGs can enhance the performance of LLMs in complex reasoning tasks while maintaining parameter efficiency.