PCA与联合近邻权值协同过滤算法研究

Collaborative Filtering Recommendation Algorithm Based on PCA with Jointly Derived Neighborhood Interpolation Weights

协同过滤推荐算法主要是通过学习用户对商品过去所作出的偏好行为来为用户作出推荐,也就是协同过滤算法会对用户评分矩阵进行用户行为偏好学习,从而为用户作出相应的推荐。但是,由于用户评分矩阵具有极大的稀疏性,稀疏性会影响推荐算法的推荐结果.针对评分矩阵的稀疏性问题,文章利用主成分分析法,对用户原始评分矩阵首先进行降维处理,将原始评分矩阵转换到主成分空间上,缓解了评分矩阵的稀疏性,同时也降低了运算的时间复杂度.利用MovieLens数据库对算法进行了实验并和联合近邻权值算法进行了比较.结果表明,本文算法有较高的准确度和运行效率.

Collaborative filtering recommendation algorithm is mainly used to make recommendations for the users by studying the preference behavior of users in the past. In the recommendation system, the recommendation algorithm make recommendation for users mainly through learning the user rating matrix. However, the users rating matrix has such great sparsity that affect the result of recommendation algorithm. This paper uses PCA to reduce the dimension of the original user rating matrix,and the original score matrix is transformed into the principal component space to alleviate the sparsity of the score matrix and the time complexity of the algorithm is reduced.By using the MovieLens database, the experiments are carried out and compared with the combined nearest neighbor algorithm. The results show that the algorithm has higher accuracy and running efficiency.