基于结构相似性的协同过滤推荐算法

Collaborative Filtering Recommendation Algorithm Based on Structure Similarity

评分向量的高维、稀疏,使得传统相似性度量方法的准确性较差.提出一种新的相似性计算方法—两阶段相似性计算算法.首先定义评分差异和差异确定度,得到用户偏好相似性;然后根据偏好相似性计算用户间的结构相似性,使用结构相似性对用户初始相似关系进行修正,使相似性计算结果更加合理.将本文方法应用于协同过滤推荐,在Movie Lens数据集上进行了实验.实验结果表明,与传统的相似性度量方法相比,新方法具有更高的准确性,可以显著提高协同过滤算法的推荐质量.

Due to high-dimensional and sparse preference vectors, the accuracy of traditional similarity methods is low. In this paper, we propose a novel similarity method：two-phase similarity computation algorithm. Definitions of rating difference and difference certainty are given, and preference similarity is obtained. Based on the preference similarity,the structure similarity is computed. Structure similarity is used to modify the initial relationships between users, which makes the similarity computation more accurate. The proposed method is applied to collaborative filtering recommendation, experiments are carded out on the MovieLens dataset. The results show that ,compared to the traditional similarity methods, the proposed method is more accurate and can improve recommendation quality of collaborative filtering significantly.