TV Shows and Movies Recommender Project

Inspiration: As a developer myself, I have been binge-watching TV shows and movies. One day, I began to struggle finding something NEW to watch. I realized people around me may be on the same boat. That’s why I invented this recommender system to target the “hidden-gem shows” on mainstream platforms

Product: A content-based filtering engine that generates a list of show recommendations based on user inputs like a past show they watched, ages, genres, and production countries

Backend Programs

preprocessing.py

1. Datasets: Amazon Prime, Netflix, and AppleTV+ Movies and TV Shows in Kaggle

• All uses a CC0 License → Can use freely

2. Cleaning/Transforming Dataframes

• Drop all the duplicate and redundant columns
• Apply One-Hot Encoding + Z-score normalization on remaining columns (works well on list-based & categorical columns too)

3. Save data into the joblib library

• Optimizes performance bc we don’t have to parse them in every run

model.py

ML Model used: K-Nearest-Neighbors with Cosine Similarity

• Use Case: For datasets with structured features (e.g., genre, artist, year for music recommendation), which is a YES!
• How it Works: Finds the k most similar items based on a given title and/OR user preferences

4. Load data into the engine
5. Feed the Model with the cosine similarity of the Transformed Dataframe
6. Recommendation Generation:

• For a given title: retrieves the most similar ones from the dataset.
• Applies filtering such as min_year, genre, min_imdb to narrow results

Frontend Programs

Website Prototype (main dashboard)

1. Model Persistence: Save the backend model and data as .onnx and .json files
2. Link to backend: Serve the .onnx file in Next.js. Render the recommended TV shows/movies back to our website as cards.

Next Steps

1. Consider using a movie/TV show API so we can apply this recommender system to every movie, instead of a limited dataset

• Add functionality to the recommender: make predictions on unseen data (not limited to the existing dataset)

2. Evaluate the model using MAP@K algorithm

• Ensures users will use the high-rated products first

3. Add movie poster as the background image for each movie card (using a poster api)
4. Deploy the website to web hosting platforms such as Vercel and Netlify (so far is limited to local hosts)