融合注意力机制的广告点击率预测模型

Advertising Click-through Rate Prediction Model Based on Attention Mechanism

针对传统的点击预测模型不能精确挖掘用户历史兴趣及目标广告的关联性问题,提出一种融合注意力机制的广告点击率预测模型。首先将离散数据经过嵌入过程映射成低维连续向量,为捕捉用户点击序列后的历史兴趣,引入注意力机制,同时为了实现用户点击行为的显式和隐式特征交互,引入xDeepFM网络并对网络结构进行优化改进,使之与注意力机制相结合。实验表明,改进模型相较于传统的深度因式分解机等模型有效提升了广告点击率的预估效果。

The traditional click prediction model can not accurately mine users’historical interests and their impacts on target advertising clicks,hence,an advertising click-through rate prediction model integrating attention mechanism is proposed.The discrete data are introduced into the embedding process and mapped into low dimensional continuous vectors.In order to capture the historical interest of users after clicking sequence,attention mechanism is introduced.At the same time,in order to make the explicit and implicit interaction features of user click behavior,xDeepFM network is introduced,and the network structure is optimized and improved by combining with attention mechanism.Experiments show that the improved model proposed effectively improves the prediction effect of advertising click-through rate compared with the deep decomposition machine model.