This project implements a recommendation system using both Matrix Factorization (SVD/ALS) and a deep learning approach, Neural Collaborative Filtering (NeuMF), to predict user-item interactions. The dataset is preprocessed by binarizing ratings, and the models are trained to learn user and item representations for ranking tasks. Additionally, an LLM-based method is included to generate ratings as part of the project requirements.
The performance of the models is evaluated using Hit Rate and NDCG, focusing on ranking quality rather than raw prediction accuracy. The pipeline includes data preprocessing, model training, prediction generation, and ranking-based evaluation, with results presented in tabular form for comparison across different approaches.