search

recommenders-team / recommenders

Best Practices on Recommendation Systems
https://recommenders-team.github.io/recommenders/intro.html
MIT License
18.89k stars 3.07k forks source link

DNN algo #339

Closed loomlike closed 5 years ago

loomlike commented 5 years ago

Add DNN-based algo, e.g.:

yueguoguo commented 5 years ago

@AaronHeee is working on NCF. He is the student of the original author of NCF. Basically he will contribute a quick start and a deep dive notebook, respectively, to the repo.

WessZumino commented 5 years ago

I don't think that a standard wide deep model will work. The next thing I am going to implement is a variational auto encoder.

loomlike commented 5 years ago

@WessZumino I think it will all depend on how to encode the features. There are couple different version of codes that utilize wide-deep model on recommendation problem, e.g. Google's https://github.com/tensorflow/models/tree/master/official/wide_deep (they uses deep only for MovieLens), and Microsoft internal 'DeepRec' repo. I'm working on modeling a recommender by using both wide and deep, by encoding co-occurrence patterns as a vector. Will see if that works.

loomlike commented 5 years ago

Related to #218 FYI, Ye is working on (e)XDeepFM internally.

nikhilrj commented 5 years ago

Closing as this is a parent issue - let's track the individual children instead