WCSE

A Cost-Sensitive Ensemble Model for Click-Through Rate Prediction

Abstract— Click-Through Rate prediction is crucial to sponsored search because it can be used to influence ranking, filtering, and pricing of ads. Therefore, estimating click-through rate (CTR) precisely makes significant difference in the efficiency of advertising on the Internet. The CTR prediction can be casted as a binary classification problem (user click as positive class and don’t click as negative class) with imbalanced data because the positive class presented with very few samples but associated with a higher identification importance. In this paper, we describe a new cost-sensitive ensemble model for CTR prediction. In this model, we used cost items to denote the uneven identification importance among classes, such that the ensemble strategies can intentionally bias the learning towards classes associated with higher identification importance and eventually improve the identification performance. For feature selection, we extracted two sets of predictive features: basic features and synthetic features. Finally, we made experiments on the dataset of KDD Cup 2012-Track 2 and tested the effectiveness of our model. Experiment results demonstrate that the cost-sensitive ensemble method significantly improve the effectiveness of CTR prediction.

Index Terms— Click-Through Rate Prediction; Imbalanced data; Cost-Sensitive Ensemble Algorithm; Feature Selection

Hongjian Liu, Defeng Guo
GiantStones Information Technology co., ltd, CHINA

[Download]

Cite: Hongjian Liu, Defeng Guo, "A Cost-Sensitive Ensemble Model for Click-Through Rate Prediction," 2015 The 5th International Workshop on Computer Science and Engineering-Information Processing and Control Engineering (WCSE 2015-IPCE), pp.621-628, Moscow, Russia, April 15-17, 2015.