改进的面向数据稀疏的协同过滤推荐算法

发布时间：2018-07-24 17:45

【摘要】：用户相似性和最近邻集合是协同过滤算法中最重要的两个步骤。传统的协同过滤算法依靠用户评分计算用户相似性并寻找K个邻居作为最近邻的方法为用户产生推荐,但是在数据稀疏的情况下,仅仅依靠用户评分使得推荐效果不准确。针对以上问题,文中提出一种改进的面向数据稀疏的协同过滤推荐算法。该方法引入用户属性相似性和用户兴趣度相似性,并结合传统的用户评分相似性计算用户间的相似度,通过多次实验调整三者的权重,并且采用动态选取邻居集合的方法确定用户的最近邻,从而为用户推荐最合适的项目,增强了方法实用性,以此来缓解用户数据稀疏性问题。实验结果表明,文中方法能够充分利用用户的各类数据信息,提高了预测评分的准确性及推荐质量。
[Abstract]:User similarity and nearest neighbor set are the two most important steps in collaborative filtering algorithm. The traditional collaborative filtering algorithm relies on the user score to calculate the user similarity and find K neighbors as the nearest neighbor to produce recommendations for the user. But in the case of sparse data the recommendation effect is not accurate only depending on the user score. To solve the above problems, an improved collaborative filtering recommendation algorithm for data sparsity is proposed. This method introduces user attribute similarity and user interest similarity, and combines the traditional user score similarity to calculate the similarity between users, and adjusts their weights through many experiments. And the method of dynamically selecting neighbor set is used to determine the nearest neighbor of the user, so that the most suitable items are recommended for the user, and the practicability of the method is enhanced, so as to alleviate the problem of user data sparsity. The experimental results show that the proposed method can make full use of all kinds of user data and improve the accuracy of prediction and the quality of recommendation.
【作者单位】：陕西师范大学计算机科学学院;陕西师范大学现代教学技术教育部重点实验室;
【基金】：中央高校基本科研业务费专项资金资助项目(GK201002028;GK201101001) 陕西师范大学学习科学交叉学科培育计划资助项目
【分类号】：TP391.3

【相似文献】