🎬 Movie Recommender System

An end-to-end Machine Learning web application that builds a personalization engine to recommend relevant movies based on content metadata analysis.

📌 Project Overview

This project processes text metadata (genres, keywords, cast, crew, and overviews) from the popular TMDB 5000 Movies & Credits datasets. By combining advanced Text Preprocessing (Stemming) and Text Vectorization techniques, the model computes similarity distances between movies. A user-friendly Flask web interface allows users to pick a movie and instantly receive 5 highly related recommendations along with rich metadata profiles.

---

🛠️ Tools & Technologies Used

🔹 Core Programming & Libraries

• Python — Core development language.
• Pandas & NumPy — Used heavily for data manipulation, handling missing/duplicate rows, merging dataframes, and feature aggregation.
• Jupyter Notebook — Interactive development environment for exploratory data analysis (EDA).

🔹 Machine Learning & Natural Language Processing (NLP)

• ast.literal_eval — Used to safely parse and extract dictionary objects nested inside stringified JSON data structural columns (genres, keywords, cast, crew).
• NLTK (PorterStemmer) — Used to stem words (e.g., converting "dancing", "danced" to "danc") to prevent redundant feature creation.
• Scikit-Learn (CountVectorizer) — Used to convert text tags into numerical vector arrays using a Bag of Words approach with max_features=5000 and English stop-words filtering.
• Scikit-Learn (cosine_similarity) — Used to calculate the angular distance between vectors to establish close relational similarities.

🔹 Data Serialization & UI Backend

• Pickle — Serializes operational assets to be loaded by the application efficiently:
  • movie_dict.pkl — Main index dataframe mapped to a dictionary format.
  • similarity.pkl — Pre-calculated mathematical similarity matrix.
  • movie_details.pkl — Extracted rich features (vote_average, runtime, tagline, overview, genres) for rendering a beautiful frontend response.
• Flask — Powers the web application server logic.

---

📂 Project Structure

Recommender-System/
├── dataset/                # Raw TMDB csv dataset files
├── notebooks/              # Jupyter Notebook containing data pipeline steps
│   └── recommendation_engine.ipynb
├── models/                 # Saved pickle artifacts (.pkl files)
├── templates/              # HTML layout view files (index.html, recommend.html)
├── static/                 # CSS stylesheets, interactive scripts, images
├── app.py                  # Core Flask application script
├── requirements.txt        # System module dependencies 
└── README.md               # Project documentation

---

⚙️ Data Pipeline Summary (As Coded)

1. Data Integration: Merged the TMDB Movies and Credits dataframes matching on the title column.
2. Feature Filtering: Retained core structural components: movie_id, title, overview, genres, keywords, cast, and crew.
3. Data Cleaning: Handled structural null fields via .dropna() and validated duplicates.
4. Tokenization Transformation: Applied space-removal techniques across fields so names like "Sam Worthington" combine into unique structural tokens like "SamWorthington" (preventing model vector confusion with other distinct people named "Sam").
5. Tag Concatenation: Compressed text metadata fields into a unified 'tags' paragraph element.
6. Vector Alignment: Stemmed the unified strings using NLTK and processed them through CountVectorizer to generate an array matrix shape of (4806, 5000).

---