Louiii/RecommenderSystems

Project about context-aware recommender systems on the #NowPlaying dataset. I demonstrate two recommender system techniques. The first uses tensor factorisation (contextual modelling) to predict ratings and Determinantal Point Processes to recommend items. The second uses Singular Value Decomposition to predict ratings and post-filtering to recommend items based on context.

Context-Aware Recommender Systems Coursework

Hierarchy

.
├── data
│   ├── Context_POP_RND // data from #NowPlaying
│   ├── Context_POP_USER // data from #NowPlaying
│   ├── rdn_train_final.json // my extracted data for training
│   ├── rdn_test_final.json // my extracted data for testing
│   ├── usr_train_final.json // additional dataset to train
│   └── usr_test_final.json // additional dataset to test
├── GUI_database
│   └── users.json // the likes/dislikes for users that use the GUI
├── SVD-model
│   └── // User profiles/hidden state of the model
├── TF-model
│   └── // User profiles/hidden state of the model
├── Technique1_SVD.py // the first model I use
├── Technique2_TF.py // the second model I use
├── GUI.py // the interface for the user
├── EvaluationMetrics.py // the functions for computing performance
├── RunMetrics.py
└── WriteFinalDataSet.py // the file I used to create the final dataset

Required Packages

Use the package manager pip to install the following packages.

• numpy
• matplotlib
• tqdm
• scipy
• PySimpleGUIQt

Replace PACKAGE_NAME with the names in the list.

pip install PACKAGE_NAME

Running

Replace FILE_NAME with the name of the file you would like to run.

python FILE_NAME.py

To use the recommender systems, run 'GUI.py'. To test them run 'RunMetrics.py'. Unfortunately #NowPlaying didn't provide the names of the tracks from Spotify, to get them back one must use the Spotify API (requires an account) to retrieve them from the track ids.