Recommendation engine using collaborative filtering in Python

Question

I have developed a search engine for restaurants. I have a social network wherein users can add friends and form groups and recommend restaurants to each other. Each restaurant may serve multiple cuisines. All of this in Python.

So based on what restaurants a user has recommended, we can zero in on the kind of cuisines a user might like more. At the same time, we will know which price tier the user is more likely to explore(high-end, fast food, cafe, lounges etc)

His friends will recommend some places which will carry more weightage. There are similar non-friend users who have the recommended some the restaurants the user has recommended and some more which the user hasn't.

The end problem is to recommend restaurants to the user based on: 1) What he has recommended(Other restaurants with similar cuisines) - 50% weightage 2) What his friends have recommended(filtering restaurants which serve the cuisines the user likes the most) - 25% weightage 3) Public recommendation by 'similar' non-friend users - 25% weightage.

I am spending a lot of time reading up on Neo4j, and I think Neo4j looks promising. Apart from that I tried pysuggest, but it didn't suit the above problem. I also tried reco4j but it is a Java based solution, whereas I'm looking for a Python based solution. There is also no activity on the Reco4j community, and it is still under development.

Although I've researched quite a lot, I might be missing out on something.

I'd like to know how would you go about implementing the above solution? Could you give any use cases for the same?

Answer 1

I think you won't find any out-of-the box solution for you problem, as it is quite specific. What you could do with Neo4j is to store all your data that you use for building recommendations (users, friendships links, users' restaurants recommendations and reviews etc) and then build you recommendation engine on this data. From my experience it is quite straightforward to do once you get all your data in Neo4j (either with Cypher or Gremlin).