paulourio / bipartite-link-prediction

Bipartite Link Prediction

This project explores various methods for bipartite link prediction using Yelp's Dataset Challenge dataset. In particular, we try to predict which restaurants a particular user will review. This was done as a class project for Stanford's Social and Information Network Analysis class (CS224W). The final report is included in the repository (writeups/final_report.pdf). It requires scikit-learn, networkx, and snap.py to run.

Running

1. Place Yelp academic datasets in data/provided
2. Run dataset_maker.py to generate examples
3. Run any of the following files:

• dataset_metrics.py (prints various properties of the dataset)
• random_baseline.py (random predictions)
• random_walks.py (make predictions using unsupervised random walks)
• similarity.py (make predictions using heuristic similarity measures)
• supervised_classifier.py (make predictions using a supervised binary classifier)
• supervised_random_walks.py (make predictions using supervised random walks, see Backstrom and Leskovec, 2011)
• svd.py (make predictions using matrix factorization)

4. Use eval.py to generate model evaluation metrics.