CineMatch | Research Portfolio & Recommendation Engine

CineMatch is a high-performance, content-based movie recommender system designed as a research portfolio. It combines advanced natural language processing (NLP) with a premium user interface to demonstrate the power of vector similarity in information retrieval.

Live Demo: https://kram2006.github.io/cinematch/

---

Core Features

1. Vector Similarity Engine

• Methodology: Uses Cosine Similarity on high-dimensional movie metadata vectors.
• Tag Fusion: Combines plot overviews, genres, keywords, cast, and directors into unique textual signatures.
• 3D Manifold Visualization: Projects the 5000-D feature space into an interactive 3D manifold using Truncated SVD for latent cluster analysis.

2. Explainable AI (XAI)

• Transparency: Every recommendation includes a mathematical breakdown of why it was chosen.
• Feature Attribution: Visualizes shared metadata features between the query and recommended movies.

3. Experimental Evaluation

• Ablation Study: Benchmarks different vectorization strategies (Bag-of-Words, TF-IDF, and SBERT).
• Precision@10: Evaluates accuracy using Genre Overlap as a scientific proxy for relevance.

4. Enterprise-Grade Reliability

• API Load Balancing: Implements random round-robin rotation between multiple TMDB API keys to maximize image loading throughput.
• Resilient Fallbacks: Multi-stage poster fetching strategy (Direct ID -> Title Search -> High-end Placeholder).

---

Project Structure

├── index.html              # Main Static Site (GitHub Pages)
├── app.js                  # Client-Side Application Logic (Core Engine)
├── style.css               # Premium "Neon Tech" UI Design System
├── data/
│   ├── movies.json         # Exported Movie List (4,806 titles)
│   ├── recommendations.json # Pre-computed Top-10 Recommendations
│   ├── manifold.json       # 3D SVD Coordinates
│   └── evaluation.json     # Pre-computed Benchmark Metrics
├── app_utils.py            # API Hooks & Metadata Utilities
├── config.py               # Configuration & API Key Rotation
├── preprocess.py           # Research Pipeline (Vectorization -> SVD)
├── evaluation.py           # Experimental Benchmarking Module
├── explainability.py       # XAI Feature Attribution Module
├── export_data.py          # Pickle -> JSON Data Export Tool
├── similarity.pkl          # Pre-computed Similarity Matrix (Large)
├── movie_list.pkl          # Processed Movie Metadata Pickle
├── cv.pkl                  # Fitted Vectorizer Model
└── requirements.txt        # Python Dependencies (for pipeline execution)

---

Deployment

GitHub Pages (Live Site)

The site is deployed automatically via GitHub Pages from the main branch. Simply push to main and the site updates.

Regenerating Data (Advanced)

If you need to regenerate the recommendation data from scratch:

1. Download the TMDB 5000 Movie Dataset and place CSVs in root.
2. Run the research pipeline:

   pip install -r requirements.txt
   python preprocess.py
   python export_data.py

---

Author

K RAMA KRISHNA NARASIMHA CHOWDARY
Research Status: ONLINE

---

© 2026 CineMatch Research Project. Released under the MIT License.