Abstract: How to recommend the most interested information to users accurately from a large amount of disordered information has become an important research subject in service recommendation system. This paper propose a recommendation algorithm based on the combination of the user's interest matrix and the global preferences, which is used for personalized service recommendation. This paper firstly introduces the interest tag mechanism to form the user interest chain, and to form the user interest matrix by filling unevaluated service and complementing evaluated service on the user service rating set. Then, this paper calculates the partial similarity by the Euclidean distance of the user??s interest matrix. Lastly, this paper combines the global preference similarity based on the user cognitive difference and the global behavior difference. The algorithm can effectively integrate the user's preference information, also reduce the sparsity of the data set, and improve the recommendation accuracy. A large number of experiments on the real MovieLens 1M data set show that the proposed algorithm significantly improves the recommendation accuracy compared with the current representative recommendation algorithms.

Key words: collaborative filtering, interest chain, interest matrix, global preference, similarity

摘要： 如何从大量无序的信息中向用户准确推荐其最感兴趣的信息，是推荐系统研究领域的重要课题。为此提出一种融合用户兴趣矩阵及全局偏好的推荐算法，用于个性化服务推荐。首先，引入兴趣标签机制形成用户兴趣链，对用户服务评分集合中未评价服务进行填充，对已评价服务进行互补，从而形成用户兴趣矩阵；其次，采用兴趣矩阵的欧几里德距离进行局部相似度计算；最后，联合用户认知差异和全局行为差异形成全局偏好相似度。算法在有效融入了用户的个性化偏好信息的同时，减少了数据集稀疏性，提高了推荐的准确性。在真实的MovieLens 1M数据集上进行的大量实验表明，与当前具有代表性的推荐算法相比，算法显著提高了推荐精度。

关键词: 协同过滤, 兴趣链, 兴趣矩阵, 全局偏好, 相似度