基于特征协同交互网络的商品点击率预测

Product Click-through Rate Prediction Based on Co-Action Network

随着用户行为数据和商品数据呈指数级增长,互联网平台面临严重的信息过载问题。可以在推荐算法中引入点击率预测模型,有效提升用户的购物体验。将特征协同交互网络(Co-Action Network,CAN)与深度兴趣演化模型(Deep Interest Network,DIN)相融合,挖掘用户与商品的深层次关系,模拟真实场景下用户与商品的交互情况,得到商品点击率预估模型CAN-DIN。在公开数据集Amazon-beauty上的实验表明,与相关基模型相比,CAN-DIN算法的准确率有一定的提升,可以应用于电商推荐场景。

With the exponential growth of user behavior data and goods data,internet platforms are facing serious information overload issues.A click-through rate prediction model can be introduced into recommendation algorithms to effectively improve users'shopping experience.By integrating the Co-Action Network(CAN)with the Deep Interest Network(DIN),explore the deep relationships between users and goods,simulate the interaction between users and goods in real scenarios,and obtain the goods click-through rate prediction model CAN-DIN.Experiments on the publicly available dataset Amazon-beauty have shown that the CAN-DIN algorithm has a certain improvement in accuracy compared to relevant base models and can be applied to E-commerce recommendation scenarios.