Movie Recommender System Project | Content Based Recommender System with Heroku Deployment

Summary

The video is an end-to-end tutorial on building and deploying a content-based movie recommendation system. The instructor introduces three types of recommender systems:

• Content-based filtering recommends items with similar content to something a user selected.
• Collaborative filtering recommends items based on the preferences or behavior of similar users.
• Hybrid systems combine both approaches.

The project uses content-based filtering to recommend movies with similar metadata.

Project workflow

1. Load and explore the data

   • Use the TMDB movie metadata and credits datasets.
   • Combine the datasets using the movie title.
   • Work with fields such as genres, keywords, plot overviews, cast, crew, and movie IDs.

2. Preprocess the data

   • Keep relevant information, remove unnecessary fields, and drop rows with missing overviews.
   • Extract values from structured fields such as genres, keywords, cast, and crew.
   • Use selected cast members and the director to create movie tags.
   • Combine selected fields into a text-based tags field.
   • Remove spaces within names so each full name is treated as one term, and convert the text to lowercase.

3. Create numerical representations

   • Use CountVectorizer to convert tags into vectors, with about 5,000 features and English stop-word removal.
   • Apply stemming with NLTK to treat word variants more consistently.
   • Calculate cosine similarity between movie vectors. The instructor explains why it is preferred to simple Euclidean distance for this text-vector comparison.

4. Generate recommendations

   • Find the selected movie’s position in the dataset and rank other movies by similarity.
   • Return the five closest matches.
   • Adjust the number of recommendations as needed; the instructor notes that results can vary in quality.

5. Build the interface

   • Create a basic Streamlit app with a movie-selection control and a recommendation button.
   • Save the processed movie data and similarity matrix with pickle for use in the app.
   • Fetch movie posters using the TMDB API and display them alongside recommended titles.

6. Deploy the application

   • Prepare deployment files, including a Procfile, setup.sh, .gitignore, and a requirements file.
   • Deploy to Heroku using its command-line tools.

The tutorial also encourages adapting the same workflow to other datasets, such as books or songs, and presenting the completed system as a portfolio project.

Main speaker and source

The tutorial is presented by a single instructor for CampusX, likely Nitish Singh.