RecSys.jl
RecSys.jl copied to clipboard
abhijithch
Feature Request - Allow for Implicit Variant
So I was wondering if any work has been done on trying to support implicit values for the ALS-wr algorithm. I have noticed that people have for instance received better results when using hybrid based approaches that use an "alpha" confidence value and rely on AUC and Precision based metric instead of ALS-wr. I might consider trying to implement this myself, but it would be nice to know if any progress has been made on this issue.
I had mainly experimented only with latent factor(ALS and SVD) methods, while building this. I can get a method for implicit values working, it might be interesting to see how it performs. I understand that there are multiple ways of hybridisation, could you point me to any articles or any kind of reference for the approach you suggest?
I would be great if you could contribute, I would be happy to support if necessary.
I haven't read on many article on this subject, but I am aware of multiple open source implementation of ALS that allow for hybridization.
https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/recommendation/ALS.scala https://github.com/benfred/implicit https://github.com/lyst/lightfm https://github.com/quora/qmf
DNN implicit technique: https://github.com/fstrub95/Autoencoders_cf
Interestingly, apparently even Netflix has moved away from using explicit ratings interestingly enough. Apparently, there are variants of ALS that are used for ranking instead of rating prediction.
For example: https://github.com/guoguibing/librec/blob/master/librec/src/main/java/librec/ranking/RankALS.java
On Mon, Dec 19, 2016 at 2:06 AM, Abhijith Chandraprabhu < [email protected]> wrote:
I would be great if you could contribute, I would be happy to support if necessary.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/abhijithch/RecSys.jl/issues/38#issuecomment-267896092, or mute the thread https://github.com/notifications/unsubscribe-auth/AB9WX7X1Hv786xBwjWww-rVG2kWSZoQpks5rJizvgaJpZM4LQLgC .