NeSyGPT: A Unified Neural-Symbolic Framework for Structured Knowledge Querying with Large Language Model

---

🌈 Model Architecture

📕 Dependencies

pip install -r requirements.txt

💡 Quick start

NOTE: Please write the API key in api_key.txt before running the code. And the api_key.txt file should have the same number of keys as num_process. The default num_process in the following shell script is 5.

# WikiSQL
>> bash scripts/wikisql/wikisql_gpt3.5.sh
>> bash scripts/wikisql/wikisql_gpt4.sh

# WTQ
>> bash scripts/WTQ/WTQ_gpt3.5.sh
>> bash scripts/WTQ/WTQ_gpt4.sh

# MetaQA
>> bash scripts/MetaQA/metaqa_1hop.sh
>> bash scripts/MetaQA/metaqa_2hop.sh
>> bash scripts/MetaQA/metaqa_3hop.sh

# WebQSP
>> bash scripts/WQSP/wqsp3.5/WQSP_overall_gpt3.5.sh
>> bash scripts/WQSP/wqsp4/WQSP_overall_gpt4.sh

# CronQuestion
>> bash scripts/CronQuestion/CronQuestion_gpt3.5.sh
>> bash scripts/CronQuestion/CronQuestion_gpt4.sh

📚 Code path

👈 🔎 Click

root
├── dataset
├── evaluate
├── output
├── scripts
├── structllm
│   └── prompt_
├── api_key.txt
├── CGdata_for_CronQuestion.py
├── CGdata_for_KGQA.py
├── CGdata_for_tableqa.py
└── CGdata_for_WQSP.py

Data Path

❗NOTE: The data we provide is the data after conversion to CGdata.

👈 🔎 Click

dataset
  ├── CronQuestion_CG
  │   ├── kg_test.txt
  │   ├── kg.txt
  │   ├── qa_test.jsonl
  │   └── qa_valid.jsonl
  ├── MetaQA_CG
  ├── WikiSQL_CG
  ├── WikiSQL_TB_csv
  ├── WQSP
  ├── WTQ
  └── deal_raw.py

Evaluate Path

👈 🔎 Click

evaluate
  ├── evaluate_for_CronQuestion.py
  ├── evaluate_for_metaqa.py
  ├── evaluate_for_tableqa.py
  └── evaluate_for_wqsp.py

Prompt Path

👈 🔎 Click

prompt_
   ├── MetaQA
   │   ├── 1hop.json
   │   ├── 2hop.json
   │   └── 3hop.json
   ├── WQSP
   │   ├── WQSP_name.json
   │   └── WQSP_unname.json
   ├── CronQuestion.json
   ├── wikisql.json
   └── WTQ.json

🎯 Results

Result on TableQA dataset

Result on KGQA dataset

Result on CronQuestion dataset