SemanticParser4Graph

Overview

This project implements

• A process to convert relational databases to property graph databases.
• A process to translate SQL queries to Cypher queries by parsing, tokenizing SQL queries and converting SQL queries to Cypher queries, adapted from the TabularSemanticParsing.
• Cy-Spider: Semantic Parsing Corpus and Baseline Models for a Property Graph.

Quick Start

Requirements

The main requirements are:

• Python 3.6+
• Neo4J Community Edition

Install Dependencies

We recommend using virtual environments to run this code:

python -m virtualenv venv
source venv\bin\activate

Python packages can be installed via:

   git clone https://github.com/22842219/SemanticParser4Graph.git

   pip install torch torchvision

   python3 -m pip install -r requirements.txt
   or 
   pip install -r requirements.txt

Download Pre-process Data

Text-to-Spider benchmarks, e.g., Spider, KaggleDBQA, and BIRD. We use the Spider benchmark to illustrate the process.

Download the pre-processed data release, and unzip the folder. put the data into application/data/spider.

Note: If you would like to preprocess Spider dataset by yourself, please refer to salesforce TabularSemanticParsing

Set up Environment

1. Setting the application/config.ini file.

   • Create an application/.env file.

   GRAPH_PASSWORD=<your-neo4j-password>

   • The application that will be run, are determined in the config.ini file:

   [FILENAMES]
   root = <path-to->/SemanticParser4Graph
   benchmark = Spider
   
   neo4j_import_folder  = <path-to->/neo4j-community-4.4.11/import>
   neo4j_uri = http://localhost:7474/browser/
   neo4j_user = neo4j
   neo4j_password = <your-neo4j-password>

   • Meanwhile, please config Neo4j export path.

   cd application/ConverDB.py
   

   Set _neo4j_export_path = '<path-to->/neo4j-community-4.4.11/import' in Class ConvertDB.

2. Configure application/conf/db.ini file

   spider_path = <path-to->/SemanticParser4Graph/application/data/spider/database
   database = musical
   
   [neo4j]
   port = 7687
   host = localhost
   username = neo4j
   password = <your-neo4j-password>
   
   

3. Running Neo4j

   cd <path-to-neo4j-bin>
   ./neo4j start

4. Constructing a property graph database from any arbitrary relational database schemas directly.

   cd application/rel_db2kg
   python schema2graph.py --<benchmark_dataset-name> --cased
   

   Translate SQL queries to Cypher queries.

   cd application/rel_db2kg
   python sql2cypher.py 
   

5. Running interface

   cd application
   python interface --web_ui

CySpider

cd semantic_parser
bash experiment-text2cypher.sh 

Citation

If you find the resource in this repository helpful, please cite

@article{zhao2023rel2graph,
  title={Rel2Graph: Automated Mapping From Relational Databases to a Unified Property Knowledge Graph},
  author={Zhao, Ziyu and Liu, Wei and French, Tim and Stewart, Michael},
  journal={arXiv preprint arXiv:2310.01080},
  year={2023}
}
@inproceedings{zhao2023cyspider,
  title={CySpider: A Neural Semantic Parsing Corpus with Baseline Models for Property Graphs},
  author={Zhao, Ziyu and Liu, Wei and French, Tim and Stewart, Michael},
  booktitle={Australasian Joint Conference on Artificial Intelligence},
  pages={120--132},
  year={2023},
  organization={Springer}
}

Related Links

The web interface is adapted from UNSW SQL2Cypher: https://github.com/UNSW-database/SQL2Cypher.