在即時競標中使用設限資料預測可獲勝的價格

Predicting Winning Price in Real Time Bidding with Censored Data

吳齊軒(Wush Chi-Hsuan Wu)

指導教授 ： 陳銘憲

國立臺灣大學/電機資訊學院/電機工程學研究所/博士(2019年)

https://doi.org/10.6342/NTU201900270

摘要

無資料

關鍵字

網路廣告 ； 即時競標 ； 機器學習 ； 深度學習

並列摘要

Real-Time Bidding is currently the most popular ad auction process for online advertising.
In this study, we study how to predict the winning price of each bid from the aspect of a bidder by leveraging the machine learning and statistical methods on the bidding history.
A major challenge is that the real winning price is not observed by the bidder after losing.
We propose to utilize the idea from censored regression model, which is widely used in the survival analysis and econometrics, to derive the loss for the losing data.
Moreover, the assumption of the censored regression is violated in the real data, so we propose a model which uses the winning rate prediction to mitigate the impact of violation.
It is named as the mixture model.
Furthermore, We generalize the winning price model to incorporate the deep learning models with different distributions and propose an algorithm to learn from the historical bidding information, where the winning price are either observed or partially observed.
We study if the successful deep learning models of the click-through rate can enhance the prediction of the winning price or not.
We also study how different distributions of winning price can affect the learning results.
Experiment results show that the censored regression usually outperforms the linear regression and the proposed averaged model always outperforms the linear regression.
Experiment results also show that the deep learning models indeed boost the prediction quality when they are learned on the historical observed data.
In addition, the deep learning models on the unobserved data are improved after learning from the censored data.
Finally, we study the combination of the mixture model and the deep learning model.

並列關鍵字

Online Advertising ； Real-Time Bidding ； Machine Learning ； Deep Learning

參考文獻

[1] J. Aldrich. R.a. fisher and the making of maximum likelihood 1912-1922. Statist. Sci., 12(3):162–176, 09 1997.

[2] L. M. Ausubel and P. Milgrom. The lovely but lonely vickrey auction. In Combinatorial Auctions, chapter 1. MIT Press, 2006.

[3] R. H. Byrd, P. Lu, J. Nocedal, and C. Zhu. A limited memory algorithm for bound constrained optimization. SIAM J. Sci. Comput., 16(5):1190–1208, Sept. 1995.

[4] O. Chapelle. Modeling delayed feedback in display advertising. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’14, pages 1097–1105, New York, NY, USA, 2014. ACM.

[5] H.-T. Cheng, L. Koc, J. Harmsen, T. Shaked, T. Chandra, H. Aradhye, G. Anderson, G. Corrado, W. Chai, M. Ispir, R. Anil, Z. Haque, L. Hong, V. Jain, X. Liu, and H. Shah. Wide & deep learning for recommender systems. In Proceedings of the 1st Workshop on Deep Learning for Recommender Systems, DLRS, pages 7–10, New York, NY, USA, 2016. ACM.