basic content based recommendation example

User 255 | 4/24/2014, 8:32:28 AM

Hi everyone,

First of all, i really liked the Graphlab python notebooks. very well explained. tons of thanks. Most of the tutorials were focused on user ratings/collaborative based recommendations. I want to start with a simple content based movie recommendation for learning purpose with some 1000 movies. I have list of movies and some info about it such as genre, movie rating, and then later add other factors to it such as mood of the movie, plot of the movies, etc. Any suggestion where to start with content based recommendation using graphlab.



User 18 | 4/25/2014, 5:29:15 PM

Hi Suvir,

Thanks for the comments!

We don't have a pure content-based recommender right now. Are you wanting to do queries such as "given movie x, give me the list of movies that are most similar to x, based on the movie features"? Right now the closest of what we have is the Item Similarity recommender, which requires user-movie interaction data. Soon it will also be able to include side features of the movies and users, but it will still require having interaction data.

There might be another way to gain some insight into your data, without doing content-based retrievals. Have you tried clustering the movies? If you transform the side info into a numeric feature vector, then you can try the KMeans model in the clustering toolkit. Play with different values of k and look at the results. It might show you interesting things about your data.