基于SE模块的神经协同过滤

Neural collaborative filtering based on SE module

基于传统推荐方法对辅助信息利用不足,为优化用户、项目间内在联系挖掘有限等问题,采用将Squeeze-and-Excitation Networks结构嵌入神经协同过滤的方法提出SE-NCF模型,利用SE模块学习权重,去除不同特征中权重较低的噪声来实现特征融合,通过神经协同过滤层获得用户-项目间的线性与非线性关系,实现模型优化。通过茶评与Amazon_Food两公开数据集对同类推荐方法进行实验,实验结果表明,相比于原神经协同过滤,SE-NCF模型在两数据集中MSE指标与NDCG指标均得到改善,在茶评数据集下MSE降低10%,NDCG提升5.1%;在Amazon_Food下MSE降低4.3%,NDCG提升9.3%。

Based on the lack of utilization of auxiliary information in traditional recommendation methods,with the purpose of optimizing the problem of limited mining of the intrinsic connection between users and items,the SE-NCF model is proposed by embedding the structure of Squeeze-and-Excitation Networks into the neural collaborative filtering method,using the SE module to learn the weights and remove the noise with lower weights in different features to achieve feature fusion The model optimization is achieved by obtaining linear and nonlinear relationships between users-items through the neural collaborative filtering layer.The experimental results show that compared with the original neural collaborative filtering,the SE-NCF model has improved MSE and NDCG indicators in both datasets,with MSE reduced by 10%and NDCG improved by 5.1%in the tea review dataset and MSE reduced by 4.3%and NDCG improved by 9.3%in Amazon_Food.