LLM-enhanced Symbolic Reasoning for KBC

Introduction

This repository provides resources for paper “Large Language Model-Enhanced Symbolic Reasoning for Knowledge Base Completion”.

Dependencies

• Python 3.9.19
• transformers==4.40.1
• scikit-learn==1.4.2
• scipy==1.13.0
• torch==2.3.0
• transformers==4.40.1
• nltk==3.8.1
• sentence-transformers==3.0.0
• sentencepiece==0.2.0
• openai==1.24.0
• google-generativeai==0.7.2
• groq==0.31.0

Code Files

The dataset for UMLs/WN18RR/FB15K found here, CN100 could be found here and WD15K could be found here with the interpretability annotations. RotatE could be found here and kge.py includes part of the said implementation. The main python file is lesr.py. We include the FB15K relation mapping in fb15k_rels.csv and example commands in run_example.sh. To run LeSR, please create subdirectories data/, log/, runs/, move dataset and kge (if used) to their respective folder, and provide own LLM inference api key. To use other knowledge base data, please edit the commandline parse and add data reading in data.py.