Hi,

Just took a look at this module. I see you are building a fancy query to create the matrix of user/node counts. From what I see, you are feeding this to the recommender module.
I am thinking of takeing the idea of the smackdown module (http://drupal.org/project/smackdown) and applying it to this. In smackdown, you show a user 2 nodes and he votes for one of them... this is, as far as I can remember, text-book collaborative filtering. What would be the best way to implement it ?

Thanks

Comments

danithaca’s picture

@PGiro: you are exactly right about "building a fancy query to create the matrix of user/node counts. From what I see, you are feeding this to the recommender module."

To apply it to smackdown module, I think the PageRank algorithm might works better. I'll implement it real soon in the Recommender API module. Another alternative might goes like this. Suppose you give a user node A and node B, and the user vote for node A. Then you might want build a table, give the tuple user-nodeA a weight 1, and give user-nodeB a weight -1. Then feed the table to Recommender API, using the recommender_similarity_classical() function.

Does that answer your question?

PGiro’s picture

Status: Active » Closed (fixed)

@danithaca : yup, that's an idea. Probably the easiest way to implement it is to simply cast two votes using votingapi and then use a fancy query/view there.

Thanks !