基于组合响应预测模型的转换率预测

Conversion Rate Prediction Based on Combined Response Prediction Model

摘要: 在实时竞价广告(Real-time bidding, RTB)中，转化率是衡量广告效果的重要指标。因为实时竞价广告的转换率很低，造成了广告日志数据的稀疏性，给通过历史数据预测转换率带来了难度。本文通过构建组合响应预测模型CRPM (Combined Response Prediction Model)预测转换率，通过点击率和转换率的联立方程来消除内生性问题，并通过推荐系统中贝叶斯个性化排序算法进行了优化，解决了数据的稀疏性问题。实验结果显示，组合响应预测模型的相较于目前常用的逻辑回归预测方法对转换率的预测有着更好的预测效果。

Abstract: In real-time bidding (RTB), conversion rate is an important index to measure the advertising effect. The conversion rate of real-time bidding advertising is very low, which results in the sparsity of advertising log data and makes it difficult to predict the conversion rate through historical data. In this paper, the CRPM (Combined Response Prediction Model) is constructed to predict the conversion rate, and the simultaneous equation of click rate and conversion rate is used to eliminate the endogenous problem. The Bayesian personalized ranking algorithm in the recommendation system is optimized to solve the data sparsity problem. The experimental results show that the combined response prediction model has a better prediction effect on the conversion rate prediction than the commonly used logistic regression prediction method.