Make recommendations to users automatically based on their list

[karllhughes]

Using the OMDB API, we could get features from movies in a user's list and then search for other movies that might have those same features.

Some features might include:

• Year released
• Rating
• Genre
• Director
• Writer
• Actors
• IMDB/Metascore ratings

All these fields are available in the API, but one problem is that their API only allows searching by title and movie year.

So, if we want to search by these fields, we'll need to index a significant portion of their results in our own database, or we could explore using another data set, like this one on Kaggle: https://www.kaggle.com/tmdb/tmdb-movie-metadata

There's also another API we could try: https://developers.themoviedb.org/3/getting-started/introduction But I don't know if they offer any better searching/filtering either.

[karllhughes]

Another data source with lots of movies and ratings in it: https://grouplens.org/datasets/movielens/latest/

A list of recommendation systems (including open source ones, some in Python) we could use to help this project along: https://github.com/grahamjenson/list_of_recommender_systems/blob/master/README.md

⭐️ A complete walkthrough for building a movie recommendation system in Python: https://www.analyticsvidhya.com/blog/2016/06/quick-guide-build-recommendation-engine-python/

How recommendation systems work: https://blog.statsbot.co/recommendation-system-algorithms-ba67f39ac9a3

[karllhughes]

We decided to use IMDB's data as this is a non-commercial project anyway. Here's how it'll work:

• [x] We'll write an importer that takes in title.akas.tsv, title.basics.tsv, and title.crew.tsv and flattens and saves the data to the database. (#59)
• [x] We'll add a Django Signal that looks for the movie based on the recommendation name, and saves a movie ID to the recommendation if one is found.
• [ ] We will create an algorithm to recommend a new movie to each user every week.

[karllhughes]

Fuzzy matching SQL Query:

SELECT id, title, year, levenshtein(lower(title), lower('halloween')) as similarity
FROM "moviesImporter_movie"
WHERE levenshtein(lower(title), lower('halloween')) < (length('halloween') / 4)
order by levenshtein(lower(title), lower('halloween')) asc, year desc;

[karllhughes]

The last step is making a recommendation algorithm. Looks like Surprise might be helpful: http://surpriselib.com/