Dear list, I have some questions regarding collaborative filtering. Although they are not specific to Spark, I hope the folks in this community might be able to help me somehow.
We are looking for a simple way how to recommend users to other users, i.e., how to recommend new friends. Do you have any experience in using collaborative filtering (MatrixFactorization) to recommend users instead of products? Are there any caveats we should be aware of or can we directly apply the method? We considered using the similarity of users (based on the sets of common friends) to suggest new friends, but (1) iterating over the whole set of users sounded inefficient and (2) we are not sure the intersections between the friend-sets is sufficiently large/diverse to yield a personalized friendship recommendation. Would MatrixFactorization be more efficient? Would it yield somehow better results due to the latent factors? Any experiences on that? Finally, our users are connected with binary values (like or dislike). Is such information sufficient to feed into the algorithm, or does the algorithm require a score from 1 to N for explicit feedback or a number of occurrences, visits, messages exchanged, etc for implicit feedback? I would be very grateful about any pointers. Cheers, Diogo PS: I know there are many many papers on these topics, but I am first trying to collect evidence that this is the right direction for us. --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
