中文摘要：

协同过滤推荐算法是应用最广泛、最成功的推荐算法之一,该算法的核心是计算用户或项目相似度矩阵.首先分析了经典的相似度度量方法存在的缺陷,即在数据稀疏时会严重影响推荐结果.针对上述问题,提出一种基于用户间的共同评分数量及评分差异度的相似度度量方法,可以缓解数据稀疏对推荐结果的影响.选择Movie Lens站点提供的著名电影评分集作为实验数据并采用五折交叉法选取测试数据,分别将本算法和基于项目的协同过滤推荐算法、基于用户的协同过滤推荐算法进行对比,结果显示：采用新相似度所得到的推荐结果在一定程度上要优于上述2种经典相似度度量方法所得到的推荐结果.

英文摘要：

Collaborative filtering recommendation algorithm is one of the most widely used and successful rec-ommendation algorithms, the core of the algorithm is to compute the similarity matrix. This paper analyzes defects ofthe classical similarity measure at first, if the data is sparse using the classic method will seriously affect the recom-mended results. This paper proposes a new similarity measure which based on users＇ difference degrees and Quanti-ties of common scores. This new method can mitigate the impact of the sparse data on the recommend results. Choos-ing movies＇ score set as the experimental data and selecting test data by half of the cross method. Classical collabor-ative filtering recommendation algorithm includes item-based collaborative filtering recommendation algorithm anduser-based collaborative filtering recommendation algorithm. The results showed new algorithm can achieve moreaccurate recommended results than classical algorithm.