MOVIE_LENS 🎥

Content-Based Movie Recommendation System

Features

• Content-based filtering using movie metadata (genre, director, actors, keywords, etc.)
• Text vectorization using CountVectorizer
• Performance enhancement with TfidfVectorizer
• Similarity scoring using cosine similarity
• Intelligent recommendations based on contextual similarity
• Lightweight, fast, and easy to deploy

---

Tech Stack and Tools

• Python
• Pandas
• Numpy
• Scikit-learn
  • CountVectorizer
  • TfidfTransformer or TfidfVectorizer
  • cosine_similarity
• Streamlit for building the frontend

---

Concept Overview

Content-Based Filtering

This system recommends movies by comparing the content of items (e.g., genres, keywords, cast) rather than relying on user ratings.

CountVectorizer

Transforms textual movie metadata into a matrix of token counts, providing a base-level numerical representation.

TF-IDF Enhancement

Replaces the CountVectorizer with TF-IDF, which reduces the importance of commonly used words and emphasizes unique terms, improving recommendation quality.

Context-Based Learning

Uses semantic similarity between movies based on metadata context (like plot keywords, genres, cast) to suggest similar movies. This includes:

• Weighted importance of certain features (e.g., genre > cast)
• Real-time similarity computation using cosine distance

---

How It Works?

1. Data Preprocessing

   • Load and clean the dataset (CSV).
   • Combine relevant text fields into a single "soup" of content.

2. Feature Extraction

   • Apply CountVectorizer to transform text into vectors.
   • Then, apply TfidfVectorizer to enhance the vectors.

3. Similarity Matrix

   • Compute the cosine similarity matrix between all movies.

4. Recommendation Engine

   • Given a movie title, fetch top N similar movies based on similarity score.

---

Thank you,

Disha Basu Krishnendu Roy