MovieLens Content-Based Recommendation System

This project implements a Content-Based Recommendation System using the MovieLens dataset. It incorporates grid search for hyperparameter tuning and uses advanced techniques like embeddings, cosine similarity, and pre-trained language models to provide movie recommendations and predict ratings.

Features

• Content-Based Recommendation: Combines movie tags and plot summaries to create a comprehensive feature set for recommendations.
• Embedding-Based Similarity: Utilizes pre-trained models to generate movie embeddings for accurate similarity computation.
• MAE Calculation: Evaluates the accuracy of predicted ratings using Mean Absolute Error.
• Top-N Recommendations: Provides top-N movie recommendations for users and calculates hit ratios.
• GPU Acceleration: Uses PyTorch for efficient computations on GPU.
• Hyperparameter Tuning: Implements grid search to optimize parameters like embedding size, learning rate, and regularization.

Dataset

The MovieLens dataset is sourced from GroupLens. It includes various files like ratings.csv, movies.csv, tags.csv, links.csv OMDb API used to fetch plots.csv

Usage

1. Preprocessing

• Dataset Preparation: Merge tags and plot summaries into a single feature column.
• Cleaning: Normalize and clean text data.
• Embedding Generation: Create movie embeddings using a pre-trained language model.

2. Model Training and Evaluation

• Grid Search: Optimize hyperparameters for the recommendation system.
• Rating Prediction: Predict user ratings for movies and evaluate using Mean Absolute Error (MAE).
• Recommendation Evaluation: Generate Top-N recommendations and calculate the hit ratio.

3. Running the Notebooks

• Grid Search Notebook: Tune hyperparameters for optimal performance.
• Content-Based Recommendation Notebook: Train, evaluate, and generate recommendations.

jupyter notebook grid_search.ipynb
jupyter notebook Content_Based_Recommendation.ipynb

Results

• MAE: Achieved an average MAE of 0.66 on the rating prediction task.
• Hit Ratio: Evaluated hit ratios for Top-N recommendations (N=10) with a robust evaluation setup.

Directory Structure

movielens-recommender/
├── data/                     # Place MovieLens dataset files here
├── grid_search.ipynb         # Hyperparameter tuning with grid search
├── Content_Based_Recommendation.ipynb  # Content-based recommendation system
├── requirements.txt          # Python dependencies
├── README.md                 # Project description
└── plots.csv                 # Plot summaries of all the movies 

Dependencies

• Python 3.8+
• Pandas, NumPy
• PyTorch
• SentenceTransformers
• scikit-learn
• tqdm