一种基于参数化重排名提高多样性的推荐方法

摘要

针对传统个性化推荐算法追求提升推荐准确率而忽略总体多样性,从而导致用户满意度降低的问题,提出一种基于参数化重排名提高多样性的推荐方法.用户历史评分尺度决定了用户的偏好相关系数,该方法据此改进了传统重排名算法中偏好相关系数的计算方法.使用参数化的计算方法获得用户排名阈值,以实现推荐准确性与多样性的有效平衡,提高了推荐的质量需求.实验结果表明,该方法在推荐精确度都降低的情况下,较其他算法能够显著增加推荐的总体多样性,提高推荐质量.%Aimimg at the problem of overemphasizing recommendation accuracy and ignoring the aggregate diversity in the traditional collaborative filtering recommendation algorithm, a method for improving recommendation diversity based on parameterized re-ranking is presented in this paper.User history ranking size decides users' preference coefficient.This paper improves the traditional re-ranking method to calculate the coefficient of users' preferences in the traditional re-ranking method by using the users' past preferences.The threshold of users' ranking is calculated by parameters, and the balance between recommendation accuracy and diversity is realized.Finally, experiment results demonstrate that the approach proposed in this paper can significantly enhance the aggregate recommendation diversity and promote the users' satisfaction compared with other similar recommendation algorithms,while reducing the recommendation accuracy.