Getting top users for an item

[chakra1166]

We have an use case where we need to get top n users for an item. We have an trained lightfm model, can we use the scores for a particular user and item using predict function and sort users based on that?

[maciejkula]

Yes.

[ayeright]

Hello @chakra1166 did it work for you? I have tried to use the scores to recommend users for a particular item and the results are extremely poor. The other way round works well.

@maciejkula when training with WARP loss would you expect the predictions to work both ways or only for ranking items given a user?