现存的广告点击率预估模型提取的特征维数较多,数据量较大,使得传统平台在应用时压力大,反应时间较长.针对这一问题,提出梯度提升决策树与因子分解机相结合的广告点击率预估模型,将基础特征库里的连续特征离散化,利用梯度提升决策树对输入特征进行非线性转化,利用Hadoop大数据平台进行分布式训练,高效快速地提取出高层特征,利用因子分解机融合模型解决不均衡分类问题,利用AUC指标对模型进行评估,与常用广告点击率预估模型进行对比.实验结果表明,大数据平台以及并行化的应用使特征提取更加高效,模型解决了分类不均问题,具有更好的广告点击率预估效果.%Existing advertising click-through-rate prediction models extract more feature dimension and the amount of data is large.So the application of the traditional platform is under pressure and reaction time is longer than others.To solve this problem,a click-through-rate prediction model combining gradient boosted decision tree and factor decomposition machine was pre-sented.Continuous feature of basic feature library was discretized,and input feature was transformed nonlinearly,also Hadoop big data platform was used to extract the high level features with high efficiency.Factor decomposition machine fusion model was used to solve the imbalanced classification problem.AUC index was used to evaluate the model,whose results were compared with that of common click-through-rate prediction model.Results of experiments show that big data platform and paralle-lization application are more efficient to implement feature extraction,and the model solves the problem of imbalanced classification.Besides it has better effects of click-through-rate prediction.
展开▼
