Skip to content

Latest commit

 

History

History
68 lines (48 loc) · 3.12 KB

README.md

File metadata and controls

68 lines (48 loc) · 3.12 KB
Raw

Movie Recommendation System

This project is a Content-Based Movie Recommendation System that uses movie metadata from the TMDB 5000 dataset to recommend movies similar to a given title. The recommendation engine is built using techniques such as text preprocessing, stemming, and cosine similarity. Try out here: https://cinemarecommendation.streamlit.app/

Dataset

The dataset used for this project includes:

  • tmdb_5000_movies.csv: Contains various details about movies like genres, keywords, overview, etc.
  • tmdb_5000_credits.csv: Includes information about the cast and crew.

Project Workflow

  1. Data Preprocessing:

    • Merged the movies and credits datasets on the id column to create a single DataFrame (new_movies).
    • Dropped unnecessary columns, keeping only the ones relevant for recommendations (id, original_title, genres, keywords, cast, crew, overview).

  2. Feature Engineering:

    • Combined overview, genres, keywords, cast, and crew columns into a single tags column, which serves as the main input for the recommendation model.
    • Used the following steps to populate the tags column:
      • Extracted genres, keywords, and cast/crew names as lists.
      • Limited the cast to the top three actors and extracted only the director’s name.
      • Removed spaces in multi-word names to avoid confusion during training.

  3. Text Vectorization:

    • Transformed the tags column into a matrix of token counts using the CountVectorizer with a maximum of 5000 features and English stop words removed.
    • Applied stemming using NLTK's PorterStemmer to reduce similar words to their root form.

  4. Cosine Similarity Calculation:

    • Calculated cosine similarity between all movies based on the vectorized tags, allowing us to measure the similarity between movies.

  5. Recommendation Function:

    • Built a recommend() function that, given a movie title, retrieves the top 5 most similar movies based on cosine similarity.

  6. Saving Model and Data:

    • Used the pickle library to save the processed movie details and similarity matrix for easy reuse.

Example Usage

The project includes an example for generating recommendations. To use the recommendation function, load the movie_list.pkl and similarity.pkl files and call recommend() with a movie title:

import pickle

# Load files
movies_df = pickle.load(open('movie_list.pkl', 'rb'))
similarity = pickle.load(open('similarity.pkl', 'rb'))

# Example: Recommend movies similar to "Baby's Day Out"
recommend("Baby's Day Out")

Project Setup

  1. Install the required libraries: 
    pip install numpy pandas matplotlib seaborn sklearn nltk
  2. Download tmdb_5000_movies.csv and tmdb_5000_credits.csv datasets.
  3. Run the notebook or script provided to process data, train the model, and generate recommendations.

Dependencies

  • Python 3.x
  • Libraries: numpy, pandas, matplotlib, seaborn, scikit-learn, nltk, pickle

Acknowledgements

Data used in this project is from The Movie Database (TMDB).