基于标签和评分差值信息熵的协同过滤算法

A Collaborative Filtering Algorithm Based on Entropy of Label and Score Difference Information Entropy

针对传统协同过滤算法存在的数据稀疏性问题,文中分析了用户标签兴趣实时变化导致的推荐准确度下降以及用户评分分散程度大等问题,提出了一种基于用户兴趣标签和评分差值信息熵的协同过滤算法。利用用户向量标签和兴趣遗忘函数计算用户实时性兴趣相似度;再通过改进后的Jaccard函数对用户间评分差值信息熵进行加权,得出用户评分相似度;最后结合用户实时性兴趣相似度和差值权重信息熵得到目标用户的相似用户集,为目标用户推荐Top-N的相似度用户。实验结果表明,改进后的协同过滤算法能有效的提高推荐精度,同时在一定程度上解决稀疏性问题。

Aiming at the sparseness of data in the traditional cooperative filtering algorithm,this paper analyzes the problems such as the decrease of recommendation accuracy and the degree of user scoring caused by real-time changes in user tag interest,and proposes a collaborative filtering based on user interest tag and score difference information entropy algorithm. In this case,the user＇s real-time interest similarity is calculated by using the user vector tag and the interest forgetting function. Then,the improved score information entropy is weighted by the improved Jaccard function,and the similarity of the user＇s score is obtained. Finally,And the difference weight information entropy to get the similar user set of the target user,for the target user to recommend Top-N similarity of the user. The experimental results show that the improved collaborative filtering algorithm can effectively improve the recommendation accuracy and solve the sparseness problem to a certain extent.