Paper for the CTR prediction with the influence of neighbor advertisement
wjbianjason opened this issue · 10 comments
As we can see, in real industry environment, an advertisement is always surrounded with other advertisements, therefore, this advertisement's CTR is influenced by other advertisement's info. But, when we construct CTR model, we usually regard it as a single point prediction. So I want to know is there some papers to settle this question, or how to define this problem, I can search some related paper. Thanks.
good questions but no related open datasets
It's an interesting question, but to some extent I think it is still a "sequential CTR predication" problem. I think you can search with keyword CTR + RNN.
You may read this paper:
Managing Risk of Bidding in Display Advertising by Haifeng Zhang et al. WSDM 2017.
Outbrain's Kaggle competition could be interesting in this regard: https://www.kaggle.com/c/outbrain-click-prediction. They show multiple recommendations (stories not ads) to a user on the same page and one has to sort the recommendations by click likelihood. Obviously, the CTR of one item is influenced by the other items in consideration. Some participants came up with features and models to use this information. @wjbianjason The dataset could be interesting for you, but also the approaches published by the kagglers.
Last answer remind me of a SIGIR 17 paper (Evaluating Web Search with a Bejeweled Player Model)
@IPL thanks for your answer, but i don't think it is a sequential problem. because in this situation, all the items are displayed at the same time.
@rk2900 thanks, i'll read it
@wsteitz thanks for your detailed answer, I‘ll follow the competition and study the participants' methods
@yufengwhy thanks, i'll read it, If i get some gain, I'll write here
good topic