2

I understand Myrrix's support for User > Item-based collaborative filtering-style, which will work well for me; but I also need to support content-based recommendations for Items, using a custom similarity algorithm. So if a user selects item X, they will also be able to see the n-most similar items, irrespective of any ratings.

That algorithm will compare Items based upon their intrinsic characteristics and attributes, and I can easily supply that algorithm in Java, but is this supported in Myrrix?

Andrew Regan
  • 5,087
  • 6
  • 37
  • 73

1 Answers1

1

In an indirect way, yes. You can add 'tags' as if they are users and items -- that is, a user tag is like an item that the user interacts with. This provides a way to inject information like user attributes, and vice versa for items. Under the hood, these work just like actual users and items in the algorithm. See the setTag() method and API methods.

Sean Owen
  • 66,182
  • 23
  • 141
  • 173
  • Ah, interesting, I think I understand. So Item X with tags A,B,C would be judged slightly similar to Y with C,D,E, and more similar to Z with A,B,D, and preferences would also be in the mix? If so, I guess I don't need to know precisely how you calculate that. I had been looking at something like: sim(A,B) = sqrt( num shared tags / num shared + num unshared) – Andrew Regan Jul 25 '13 at 13:21
  • 1
    That's right -- well, if you imagine the tags as "fake users" you will get the idea completely. That's how they act. – Sean Owen Jul 25 '13 at 13:42