摘要
协同过滤算法是经典的个性化推荐算法,其中相似度度量方法直接影响推荐系统的准确率。针对用户评分极端稀疏情况下传统相似度度量方法均存在各自的弊端,导致推荐系统的推荐精度不高问题,提出了一种基于互信息的项目协同过滤推荐算法。该算法将互信息作为相似度度量方法,不仅考虑了变量之间的线性或非线性相关性,而且还能挖掘变量之间的相关性强弱。另外,由于共同评分的项目用户数很少,在互信息方法基础上引入了一个平滑系数因子,来缓解共同评分过少项目之间相似性度量不准确问题。最后,在公开的MovieLens、Jester两个数据集上进行了大量对比实验。实验结果表明,新算法能在一定程度上提高推荐系统的预测准确率,并能缓解数据稀疏性问题。
Collaborative filtering algorithm is a classic personalized recommendation algorithm, in which the similarity measurement method directly affects the accuracy of the recommendation system. There are disadvantages in the traditional methods of similarity measurement under the extreme sparseness of user ratings, leading to the recommendation accuracy of recommendation system is not high, and a items collaborative filtering recommendation algorithm based on mutual information is proposed. Mutual information is used as similarity measurement method. It not only considers the linear or nonlinear correlation between variables, but also mining the correlation between variables. In addition, due to the small number of users who join in score items, a smooth coefficient factor based on the mutual information method is introduced to alleviate the inaccuracy of the similarity measurement between items with too few common scores. Finally, a large number of comparative experiments on the public two data sets ( MovieLens, Jester) are conducted. The experimental results show that the new algorithm can improve the prediction accuracy of the recommendation system to a certain extent, and can alleviate the data sparseness problem.
作者
郑诚
章金平
徐启南
ZHENG Cheng;ZHANG Jin-ping;XU Qi-nan(Key Laboratory of ICSP Ministry of Education, Anhui University, Hefei 230601, China;School of Computer Science and Technology, Anhui University, Hefei 230601, China)
出处
《测控技术》
2019年第4期41-44,49,共5页
Measurement & Control Technology
关键词
协同过滤
互信息
平滑系数
推荐
collaborative filtering
mutual information
smoothing coefficient
recommendation