CineMatch

Live Demo

A hybrid movie recommendation engine built on Hadoop MapReduce using Python (mrjob), implementing three recommendation approaches:

• Content-Based Filtering — TF-IDF weighted genre similarity
• Collaborative Filtering — user co-occurrence patterns
• Hybrid — combines both scores by multiplication

Architecture

ratings.csv --> Content-Based (mrjob, 2 MRSteps) --> contentout.txt --+
                                                                      |
ratings.csv --> Co-occurrence (mrjob, 2 MRSteps) --> cooccur.txt      +--> Hybrid (mrjob) --> hybridout.txt
                     |                                                |
                     v                                                |
               Scoring (mrjob, 2 MRSteps) --> collaborativeout.txt ---+

Requirements

• Python 3.7+
• mrjob
• Hadoop cluster (HDFS + YARN)

pip install mrjob

Dataset

Uses the MovieLens 100K dataset in dataset/:

• movies.csv — 9,125 movies with genres
• ratings.csv — 100,004 ratings from 671 users

Running

Full Pipeline (Hadoop)

bash Codes/run_all.sh

Individual Jobs

# Content-based
python Codes/content_based.py -r hadoop --movies dataset/movies.csv dataset/ratings.csv -o output/content/

# Collaborative — Phase 1: Co-occurrence
python Codes/cooccurrence.py -r hadoop dataset/ratings.csv -o output/cooccur/

# Collaborative — Phase 2: Scoring
python Codes/collaborative.py -r hadoop \
    --movies dataset/movies.csv \
    --cooccurrence output/cooccur/part-00000 \
    dataset/ratings.csv -o output/collab/

# Hybrid
python Codes/hybrid.py -r hadoop \
    --content output/content/part-00000 \
    --collab output/collab/part-00000 \
    dataset/ratings.csv -o output/hybrid/

Local Testing (no Hadoop required)

Replace -r hadoop with -r local to run on your machine:

python Codes/content_based.py -r local --movies dataset/movies.csv dataset/ratings.csv -o output/content/

Output

Results are written to output/:

• contentout.txt — content-based recommendations
• collaborativeout.txt — collaborative filtering recommendations
• hybridout.txt — hybrid recommendations

Format (tab-separated): userId\tmovieId\tmovieTitle\tscore (sorted by userId ascending, score descending)

Website

An interactive cinema-themed web interface for exploring recommendations.

Live: https://hpk369.github.io/CineMatch/

python Codes/generate_web_data.py        # Generate JSON from output files
cd docs && python -m http.server         # Serve locally

Tech Stack

• Processing: Python, mrjob, Hadoop MapReduce (Streaming)
• Data: MovieLens 100K (CSV)
• Algorithms: TF-IDF, Co-occurrence Matrix, Score Fusion
• Web: HTML5, CSS3, JavaScript