Fidelio, pipeline hybrid movie recommender system

A pipeline movie recommendation system developed as part of the Machine Learning (ML) course.

The Pipeline Architecture

This project implements one hybrid methodology to solve the cold-start problem and improve recommendation accuracy:

🔹 Pipeline: The Gradient Boosted Hybrid

A feature-based machine learning approach that treats recommendation as a regression/ranking problem:

1. Content Encoding: Metadata is encoded into numerical features.
2. Collaborative Features: User and Item embeddings/vectors are used as input features.
3. Gradient Boosting Machine (GBM): A Tree-based model (Hist Gradient Boosting) is trained on these combined features to predict the specific user rating.

Data Sources & Acknowledgments

This project utilizes a custom dataset constructed by merging user ratings with rich movie metadata collected from multiple sources:

• MovieLens Dataset
  User–item ratings are derived from the MovieLens datasets provided by GroupLens Research.
  These ratings form the backbone of the collaborative filtering pipeline.

  • https://grouplens.org/datasets/movielens/

• The Movie Database (TMDB) API
  Rich movie metadata (TMDB IDs, cast, crew, directors, genres) was collected programmatically via the TMDB API and used primarily for content-based modeling.

  • https://developers.themoviedb.org/3

• Rotten Tomatoes Reviews Dataset
  A large-scale dataset of critic and audience reviews scraped from Rotten Tomatoes, used to enrich item representations and provide additional signals for cold-start movies and sentiment-aware modeling.

  • https://www.kaggle.com/datasets/priyamchoksi/rotten-tomato-movie-reviews-1-44m-rows

Trained Model

The complete documentation and the setup tutorial are not included in this repository.

You can see everything on Github: https://github.com/jacob14047/Fidelio