Recommendation Engine

Using scraped data from Metacritic and IMDB, this model will take a TV Show as input and return 10 others that are recommended based on that show. Written in Python using mostly selectolax, scikit-learn, fuzzywuzzy, nltk, and pandas.

Model

The underlying model uses the following features (each weighted differently):

• Genres
• Plot
• Synopsis
• Cast
• Production company
• Keywords (describing the show)
• Number of seasons
• Episode runtime
• IMDB Rating
• Metacritic score
• Metacritic user score

Using these, the model finds pairwise cosine_similarities between every TV Show in the database. Combining the top 30 most similar with a weighted average of IMDB and metacritic scores gives an overall recommendation score.

Demo

You can see a demo of prediction on my website.

Limitations

Metacritic scores are based on the first season (this is why metascores carry a lower weight). In the future, we need to scrape data for the entire show or the average of all seasons.

Setup

metacritic_scraper.py must run before imdb_scraper. Following this, the training and predicting can be done.

You can either run the main file in a terminal or run this as a flask webserver using flask_api.