Movie Recommendation API using Content Based Filtering

Technologies Used

• Docker
• Elastic Search
• FastAPI
• SBERT
• NLTK

---

Steps to run

1. Make a virtual environment with python 3.10.2

   python3.10.2 -m venv {env_name}

2. Run virtual env and install dependencies

   ./{env_name}/Scripts/activate // for windows
   source {env_name}/bin/activate // for linux
   
   pip install -r requirements.txt

3. Create a .secrets.toml file to store environment variables. Add following variables

   elastic_username = "{username}"
   elastic_password = "{password}"

4. Create docker containers for elastic search and kibana.

   docker-compose up

5. Index data in elastic search.

   python ElasticIndexer.py

6. Run FastAPI server

   uvicorn app:app

7. Go to http://localhost:8000/docs to test the API.

---

How It Works

Content Based Filtering is a type of recommendation generation system that used metadata of an item to find other similar items. The metadata for a movie would be its title, plot, cast, directors, genre, and so on. The metadata is then used to create a single document by combing them. The document in then encoded to a vector which represents that movie in the latent space. To generate recommendations for user that has liked a particular movie, the similarity is computed between that movie vector with all other movie vector. Most similar movie vectors can then be recommended to the users.

Outputs

• Input: The Lion King

• Input: Space Aliens