NuoLeiNYU / GNN-acquisition-prediction

Using a Global Corporations Knowledge Graph to Predict Mergers and Acquisitions

Project Description

Developed a Knowledge GraphDB with ~50k global corporations. Data extracted using SPARQL to query WikiData, WikiMedia’s massive Semantic Triplestore (~91m triples). Example Schema: (Company)--[size, location, industry]--(Company). Goal is to predict Mergers and Acquisitions using link prediction ML algorithms from metadata and graph embeddings (Node2Vec). Example prediction: (Company)--[ACQUIRED]--(Company)

Project Status

• Write Wikidata Sparql Queries for global corporations and corp-to-corp acquisition relationships
• Build/Run extraction and cleaning of data using Wikidata API
• Build/Run ETL process to prepare CSVs for Neo4j GraphDB Node and Edge Ingestion
• Run actual ingestion with pre-processed data
• Inspect and verify ingestion
• Peform Graph analytics exploratory data analysis (still important for ML even in Graph!)
• Test Graph Database embedding models and research more graph native link prediction algos
• Formalize ML approach and record performance
• Clean repo and publish project

Project Tools

• Python
  • sparql_config
  • Pandas
  • JSON
  • MultiThreading
  • NetworkX
  • TensorFlow
  • Sklearn
  • Plotly
• Neo4j
  • Cypher
• Docker base images
  • Neo4j
• Sparql
• Wikidata

Publication

• Predicting Mergers and Acquisitions using Graph-based Deep Learning

Installation and Running (not ready yet)

git clone git.repo.com

python installation