In this project, we built a graph embedding model that extracts precise node embeddings without using other node features. We used the idea that ‘Positive node pairs have a higher dot product between their embeddings compared to negative node pairs’
For this project, we utilized the Cora dataset. Cora dataset consists of a citation network of scientific publications. Each publication (node) is connected by edges (citations) to other publications. The nodes are also labeled with classes that represent the topic of the paper.
- 'BPR_link_pred.ipynb' is the code for our BPR-based model.
- 'GAT_link_pred.ipynb' is the code for two baseline models.
- 'GAT' is a static graph attention network.
- 'GATv2' is a dynamic graph attention network.
- We used the basic T4 GPU in Colab.
- For our project, we have studied and referenced this paper.
- <BPR: Bayesian Personalized Ranking from Implicit Feedback> (2012) https://arxiv.org/abs/1205.2618
- I worked with two colleagues at KAIST for this project (Hyeonseop Jung and Seokwon Han)
- This project was a term project for CS471: Graph Machine Learning and Mining at KAIST