TY - GEN
T1 - The prediction of CTR based on model fusion theory
AU - Chen, Jiehao
AU - Wang, Shuliang
AU - Zhao, Ziqian
AU - Shi, Jiyun
N1 - Publisher Copyright:
© Springer Nature Singapore Pte Ltd. 2017.
PY - 2017
Y1 - 2017
N2 - Online advertising makes it possible to show different ads to different customer groups according to their own characteristics, which will definitely prove the efficiency of ads, and we manage to accurate advertising by predicting the CTR of ads based on varieties of algorithm and models. This essay presented a kind of merged model of GBDT and LR, whose accuracy doesn’t heavily depend on the effect of building features artificially. In the GBDT part of the new model, the ways to build the decision trees made it possible to recognize the effective combination of features, on the other hand, the LR part of model makes it possible to deal with large amount of data. At the same test condition, the new model performed better than LR at the range of 1.41% to 1.75% with the standard of MSE, AUC and Log Loss. The results of the experiment show that GBDT model did a great job on building features for LR model without much help from human, which provides a new thought to improve the current CTR prediction models.
AB - Online advertising makes it possible to show different ads to different customer groups according to their own characteristics, which will definitely prove the efficiency of ads, and we manage to accurate advertising by predicting the CTR of ads based on varieties of algorithm and models. This essay presented a kind of merged model of GBDT and LR, whose accuracy doesn’t heavily depend on the effect of building features artificially. In the GBDT part of the new model, the ways to build the decision trees made it possible to recognize the effective combination of features, on the other hand, the LR part of model makes it possible to deal with large amount of data. At the same test condition, the new model performed better than LR at the range of 1.41% to 1.75% with the standard of MSE, AUC and Log Loss. The results of the experiment show that GBDT model did a great job on building features for LR model without much help from human, which provides a new thought to improve the current CTR prediction models.
KW - CTR prediction
KW - Gradient boosting decision trees
KW - Logistic regression
KW - Model fusion
UR - http://www.scopus.com/inward/record.url?scp=85014937177&partnerID=8YFLogxK
U2 - 10.1007/978-981-10-3969-0_11
DO - 10.1007/978-981-10-3969-0_11
M3 - Conference contribution
AN - SCOPUS:85014937177
SN - 9789811039683
T3 - Communications in Computer and Information Science
SP - 90
EP - 100
BT - Geo-Spatial Knowledge and Intelligence - 4th International Conference on Geo-Informatics in Resource Management and Sustainable Ecosystem, GRMSE 2016, Revised Selected Papers
A2 - Yuan, Hanning
A2 - Geng, Jing
A2 - Bian, Fuling
PB - Springer Verlag
T2 - 4th International Conference on Geo-Informatics in Resource Management and Sustainable Ecosystem, GRMSE 2016
Y2 - 18 November 2016 through 20 November 2016
ER -