MovieLens Recommender Systems

An educational project that compares several recommendation methods using the MovieLens latest-small dataset.

Models

Model	What it does	Status
Content-based	Finds movies with similar titles, genres, and tags	Complete
Item-item kNN	Recommends from similarities in user ratings	Complete
Matrix factorization	Learns user and movie embeddings to predict ratings	Experimental
BPR	Learns to rank watched movies above unseen movies	Complete
SASRec	Uses a user's ordered history to predict the next movie	Implemented

Saved results

These results come from the saved notebook runs. The models use different objectives, so their scores should not be compared directly.

Model	Metric	Result
Global-mean baseline	Test RMSE	1.044
Item-item kNN	Test RMSE	0.860
Matrix factorization experiment	Test RMSE	0.872
BPR	Hit Rate@10	0.113
BPR	NDCG@10	0.066
BPR	MRR	0.062
SASRec	Best validation Hit@10	0.8185
SASRec	Best validation NDCG@10	0.6065

The matrix-factorization notebook includes an SVD++-style implicit-history term, but its saved evaluation does not use that term. Treat its RMSE as an experimental result rather than a valid SVD++ benchmark.

Files

• content_based_rec.ipynb - content-based recommendations
• knn_rec.ipynb - item-item collaborative filtering
• simple_rec.ipynb - matrix factorization experiment
• BPR.ipynb - Bayesian Personalized Ranking
• sequential_recommendation/SASRec.ipynb - sequential recommendations
• movies.csv, ratings.csv, tags.csv, links.csv - MovieLens data

Run the project

Create a virtual environment and install the main dependencies:

py -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install numpy pandas matplotlib scikit-learn jupyterlab ipykernel

For SASRec, also install PyTorch and RecBole:

python -m pip install torch recbole

Start JupyterLab from the repository root:

python -m jupyter lab

Dataset

This repository uses MovieLens ml-latest-small:

• 100,836 ratings
• 9,742 movies
• 610 users

The dataset was created by the GroupLens Research Group. Review the MovieLens usage terms before redistributing the data or using it commercially.