融合共同评分用户数和项目兴趣关系的推荐算法

Recommendation Algorithm Combing Number of Co-Rating Users and Interest Relationship of Items

基于项目的协同过滤系统中,传统的cosine、pearson和jaccard等相似性计算方法都只从单一的角度衡量了项目之间的相似性,导致相似性计算结果不准确。协同过滤系统中普遍存在评分数据稀疏性问题,使这一现象更为严重。针对这些问题,提出了一种融合共同评分用户数和项目兴趣关系的推荐算法。算法根据项目间的同类性和共同评分用户数改进了Pearson相似性,引入用户对项目的兴趣度,基于用户建立项目之间兴趣度向量。实验结果表明:算法提高了推荐准确度,有效地缓解了评分数据稀疏性对传统相似性计算的影响。

In the item-based collaborative fihering system, the traditional cosine, pearson, jaccard and other similarity calculation methods only measure the similarity between the items from a single point of view, resulting in inaccurate calculations of similarity. In addition, review data sparsity problem generally exists in collaborative filtering system, which makes this phenomenon more serious. To solve these problems, a recommendation algorithm combing the number of co-rating users between items and interest relationship of items is proposed. According to the similarity of items and the number of co-rating users, the algorithm improves the Pearson similarity. The algorithm intro- duces the degree of user interest for the items and establishes vector of interest among items based on the user. The experimental results show that the proposed algorithm can improve the accuracy of recommendation, and can effectively alleviate the impact of the sparse data on the traditional similarity calculation.