摘要： 在线广告是各大互联网公司的主要盈利模式之一,目的是在满足用户的实时需求时,通过竞价和预测用户点击率等方式实现利润最大化。为实现上述目标,提出一种改进的在线广告并行运算模型。应用Logistic回归模型对用户的点击率进行建模,该模型既包含关于长期历史
信息的一次模型和二次模型因子,又包含短期的上下文因子。通过贝叶斯后验分布理论对参数计算进行推导,根据Thompson采样和预先计算2种方法改进模型计算效率。实验结果表明,该模型不仅具有较高的预测准确性,而且提高了算法的收敛速度与运行效率。
关键词:
计算广告,
响应预测,
机器学习,
Logistic回归,
点击率
Abstract: Online advertising is one of the most important profit models in Internet companies,and it aims to maximize the profit according to auction and prediction of Click-through Rate(CTR) of users while satisfying users timely requests.In order to achieve that purpose,this paper proposes a Logistic regression based online advertising parallel model.It uses the Logistic regression modeling the CTR of users.The model contains long term factors including linear and quadratic factors,and short term context factor.And it infers the computation of parameters based on Bayesian posterior.The paper uses Thompson sampling and pre-computing to improve the computation efficiency of the model.Experimental results show
that,compared with related researches,the proposed model has better prediction accuracy and quicker convergence,and thus a better computation efficiency.
Key words:
computational advertising,
response prediction,
machine learning,
Logistic regression,
Click-through Rate(CTR)
中图分类号: