10
CyneMatch — Movie Recommendation System
Web movie recommendation app based on IMDB data. Content-based engine with Scikit-learn (cosine similarity), Django interface, and full Docker containerization.
Stack
What happened?
- 01
The question
Can a meaningful movie be recommended from the similarity between the movies you already like?
- 02
The problem
Recommending by raw genre is crude. Average rating ignores what actually makes you like a movie.
- 03
The hypothesis
By representing each movie with its features (genres, keywords, cast) and measuring cosine similarity, we get a “movies like this” recommendation.
- 04
The build
Content-based recommendation engine with Scikit-learn (cosine similarity), Pandas & NumPy processing, Django interface, and full Docker containerization.
- 05
The failure
The first results recommended “similar” movies nobody would have associated. The problem: some features dominated the whole computation.
- 06
The discovery
Feature normalization and weighting radically change relevance. Content is good, but you have to dose it.
- 07
The result
A Dockerized Django web app that recommends movies from IMDB data.
- 08
The next frontier
Blending content and collaborative filtering (hybrid), and seriously evaluating recommendation quality.