Our research takes a data-driven approach to recommend co-starring in a movie. We collected movie and actor data from multiple sources to construct a network, and utilized node embeddings (EGES, Node2Vec) and message passing machanisms (GCN, SEAL) to solve a link prediction task.
We conducted experiments on the collaboration network data of movie actors using four models, which include:
1) Baseline model (ML-based): only using features of the two actors and predicting with XGBoost Classifier,
2) Benchmark models (EGES, GCN): utilizing actor and network information,
3) Best model (SEAL): employing a more advanced model architecture for actor collaboration link prediction.
We used AUC as the model evaluation metric.
| ML-based | EGES | GCN | SEAL | |
|---|---|---|---|---|
| Valid | 61.1% | 59.2% | 67.2% | 84.5% |
| Test | 55.3% | 59.2% | 67.6% | 80.1% |
Below are some examples shown on a Gradio interface.
| Contributor | Work |
|---|---|
| Jih-Ming Bai | Problem Formulation, Model, Experiment and Analysis |
| Cheng-Yu Kuan | Literature Review, Gradio Demo |
| Po-Yen Chu | EGES Model and Experiment |
| Shang-Qing Su | Data Collection, Report Delivery |
| Chia-Shan Li | Data Collection, Report Delivery |