https://github.com/seatedro/bst-lstr

Long Short Term Transformer Recommendation System
https://github.com/seatedro/bst-lstr

Last synced: about 1 month ago
JSON representation

Long Short Term Transformer Recommendation System

• Host: GitHub
• URL: https://github.com/seatedro/bst-lstr
• Owner: seatedro
• License: mit
• Created: 2022-10-10T15:52:23.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-10-28T11:32:24.000Z (over 2 years ago)
• Last Synced: 2024-11-01T10:24:46.735Z (3 months ago)
• Language: Jupyter Notebook
• Size: 10.8 MB
• Stars: 2
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# bst-lstr

Long Short Term Transformer Recommendation System

### Dataset:

[MovieLens 1M dataset](https://grouplens.org/datasets/movielens/1m/)

### Folder Structure:

 ```

  bst-lstr/

  │

  ├── data/ - directory for storing train and test data

  ├── util/ - directory for additional colab notebooks

  └── saved/

      ├── models/ - trained models are saved here

      └── log/ - default logdir for tensorboard and logging output

  ```

### Work Done:

- Implemented Behavior Sequence Transformer in PyTorch

- Combine long and short sequence lengths and their respective positional embeddings and pass to the encoder.

- Predict the `target_rating` using Dense layers

### Work to be done:

- [x] Add padding for the sequences to cover more movies in the train dataset

- [ ] Test when overfitting occurs

- [ ] Adding support for GRU model for matching

- [ ] Run inference on a deployed model