融合项目嵌入表征与注意力机制的推荐算法

Recommendation algorithm based on item embedding characterization and attention mechanism

为挖掘用户行为信息的隐性反馈,提出一种融合项目嵌入表征与注意力机制的深度学习序列化推荐算法。由于项目间的相关性特征和用户行为特征对提高推荐算法准确性有重要意义,在GRU网络模型的基础上,考虑点击序列中项目间的相关性,通过word2vec将浏览项目表征为高维向量,弥补传统项目表征的不足;用停留时间优化注意力机制刻画用户兴趣偏好,提高推荐准确性。通过实例验证了所提算法的有效性,召回率达到了69.61%,分类效果优于其它模型。

To mine implicit feedback of user behavior information,a deep learning serialization recommendation algorithm was proposed,which integrated project embedding representation and attention mechanism.Because the correlation between products and user behavior characteristics are important to improve the accuracy of recommendation algorithm,based on the GRU network model,the correlation between products in click sequence was considered,and word2vec was used to represent browsing pro-ducts as high-dimensional vectors to compensate for the deficiencies of traditional product representation.Users’interest prefe-rences were described using the residence time optimization attention mechanism,improving the accuracy of recommendation.The effectiveness of the proposed algorithm is verified by an example.The recall rate reaches 69.61%,and the classification effect is better than other models.