search

wzhe06 / Ad-papers

Papers on Computational Advertising
https://github.com/wzhe06/Ad-papers
MIT License
4.23k stars 1.18k forks source link

Wide&Deep,FNN,PNN for CTR prediction #10

Closed WuLC closed 6 years ago

WuLC commented 6 years ago

I see that Wide&Deep model for CTR prediction was removed,but according to my experiment on the dataset of Criteo in the Kaggle competition, Wide&Deep performs better than FFM. I think even thought the paper of Wide&Deep deals with recommendation problem, but both recommendation and CTR prediction can be classified as ranking problem.

Besides, FNN and PNN from Weinan Zhang are added, cause I think it is improvement on the Deep part of Wide&Deep. But the order of their publishment is FNN->Wide&Deep->PNN.

Also, I think many papers published later borrow many ideas from these three papers,like DeepFM.etc

wzhe06 commented 6 years ago

Exactly, Wide&Deep model isn't removed, just is moved to the recommendation doc. But I think you are right, we should maintain a complete list on the CTR domain. I will remove the whole reco doc and build another git repo in the future. There is no need to put a reco paper list in ads domain.