为了进一步提高点击率预估模型的预估能力,提出了一种融合时序信息并带有注意力机制的广告点击率预估模型AGCN(Attention GRU &Cross Network),该模型采用并行结构融合交叉网络模型和时序模型,实现广告点击率预估过程中多元特征的融合.该模型中交叉网络模型挖掘低阶特征和高阶特征信息,时序模型通过引入带有注意力机制的门控神经单元(GRU with attentional update gate,AUGRU),获取用户兴趣特征在用户长期兴趣演化过程中的贡献程度,进行用户兴趣筛选.实验表明,AGCN模型能有效提高广告点击事件的预测准确率.
An advertisement CTR model integrating temporal information and attention mechanism
In order to further improve the prediction ability of the click-through rate prediction model,this paper proposes an advertisement click-through rate prediction model AGCN(Attention GRU &Cross Network)that integrates time series information and attention mechanism.The model adopts a parallel structure to integrate the Cross Network model and the time series model.,to realize the fusion of multiple features in the process of advertising click-through rate estimation.In this model,the Cross Network model mines low-order features and high-order feature information,and the time series model acquires user interest features by introducing a gated neural unit(GRU with attentional update gate,AUGRU)with an attention mechanism.Contribution degree of user interest screening.Experiments show that the AGCN model can effectively improve the prediction accuracy of advertisement click events.
万方数据
维普
