基于神经协同过滤的个性化商品推荐方法

Personalized product recommendation method based on neural collaborative filtering

针对现有协同过滤方法对用户与商品的潜在信息挖掘不全面的问题,提出了一种基于多特征融合和外积神经协同过滤的个性化商品推荐方法。该方法分别采用多层感知器和卷积神经网络提取用户与商品之间的交互关系矩阵,充分利用拼接方法和外积运算的互补性,提高了对用户与商品关系的表征能力。利用外积神经协同过滤模型提升模型稳定性和拓展性。亚马逊公开数据集的测试结果表明,与原有单一特征的推荐模型相比,多特征融合能够有效提高商品评分预测性能,且推荐性能优于现有协同过滤方法。

The existing collaborative filtering method is not comprehensive in mining the potential information of users and commodities.Therefore,in this paper,a personalized product recommendation method based on multi-feature fusion and EPN collaborative filtering is proposed.The method uses multilayer perceptron and convolution neural network to extract the interaction matrix between users and commodities.The complementarity of splicing method and outer product operation is used to improve the representational ability between users and commodities.The stability and expansibility of the model is improved by using the external cumulant nerve collaborative filtering model.The test results of Amazon public dataset show that compared with the original single feature recommendation model,multi-feature fusion can effectively improve the performance of commodity score prediction.The recommendation performance of this method is better than the existing collaborative filtering methods.