基于四阶奇异值分解的推荐算法研究

Recommendation Algorithm Based on 4th-order Singular Value Decomposition

三阶奇异值分解推荐算法可以综合考虑用户、物品标签和物品三部分信息,挖掘三者之间的潜在关系进行推荐,然而该方法并没有引入其他方面的有效信息,如用户情感。为了考虑更多维度的信息,本文在三阶奇异值分解推荐算法的基础上,提出了一种加入用户情感信息的四阶奇异值分解推荐算法。该方法基于从评论中的emoji表情提炼出的用户情感偏好,再引入四阶张量模型,存储用户、用户情感、物品标签和物品四元组数据,应用四阶奇异值分解,从而进行个性化推荐。在某在线互联网教育的实证数据集上的实验结果表明,该方法比三阶奇异值分解推荐算法以及传统推荐算法在准确率和召回率性能指标上都有明显提升,其中进行Top-1推荐时,准确率和召回率可以达到0.513和0.339。本文的工作为移动通信端的个性化推荐提供了借鉴。

The 3rd-order singular value decomposition recommendation algorithm can comprehensively consider the three parts of information of user, tag and item, and explore the potential relationship between the three to make recommendations. However, this method does not introduce any other effective information, such as the user's emotion. Considering more dimension information, in this paper we propose a 4th-order singular value decomposition recommendation algorithm based on the third-order one. The method extracts user's emotional preference from the emoji expression in the commentary, introduces a 4th-order tensor model to store user, user emotion, tag, and item quad data, and applies the 4th-order singular value decomposition to make personalized recommendations. The experimental results on an empirical dataset of online internet education shows that the proposed method has a significant improvement in accuracy and recall performance than the third-order singular value decomposition recommendation algorithm and the traditional recommendation algorithms. In the Top-1 recommendation, the accuracy rate and recall rate of proposed method can reach 0.513 and 0.339. The work of this paper provides a reference for personalized recommendation of mobile.