Movie RecommenderS

Flask app based website for 2 movies recommenders:

• one based on 5 movies rating, using Collaborative Filtering NMF (non-negative matrix factorization) as rating model
• the second based on the user favourite movie via Neighborhood-based Cosine-Similarity model.

This unsupervised learning model example train on data from MovieLens Dataset 100k small dataset. The project is builted to save the input from the user, uptating the database and improving the models every 12 hours. ⏰

➡️ Repo Structure

• Data:

  • raw: from MovieLens Dataset 100k small dataset
  • prepocessed: the final dataset with the new user imput uptated every 12h.

• MICE_imputer.py -> module to generate the clean df_final.csv datset filling the missing rating using Multiple Imputation by Chained Equations: an imputation algorithm which replaces missing data values through an iterative series of predictive models, in this case a Random Forest regression model.

• Webapp:

  • app.py -> the main program to run to visualize your the website on your browser

  • recommendation_model.py -> train the NFM model based on the updated final dataset in data/preprocessed folder

  • knn_recommending_engine.py -> second model using KNN with cosine similarity distance

  • recommending_engine.py -> generate the predictions based on the NFM model and update df_final.csv file with new ratings by new users.

  • webscraper_img.py -> generate the link to the recommended movies

  • 🗂 templates/ : contains html files for the graphic visualisation

  • 🗂 static/ : contains CSS files for the webpages

Credits

This project was done in collaboration with Francesco Mari & Behzad Azarhoushang.