一种改进的基于用户聚类的协同过滤推荐算法

A novel user clustering based collaborative filtering algorithm

为克服在线视频网站中出现的数据稀疏性和推荐实时性不佳的问题,本文提出一种基于用户聚类的改进算法。首先该算法以商品属性为辅助预填充矩阵空白,然后采用初始聚类中心优化的k-means算法在矩阵上对用户进行离线聚类,将兴趣点相同的用户聚集到同一类别中,最后在线寻找目标用户最近邻并产生推荐。本文采用Movie Lens作为测试数据集,实验结果表明,本文算法可以有效缓解数据稀疏性及改善实时性,并在一定程度上提高推荐精度。

In order to solve the problems of data sparsity and poor real - time existing in online video websites, a novel user clustering based algorithm is proposed. Firstly, the rating matrix based on similarity of item attributes is prefilled. And then, k - means clustering algorithm with meliorated initial center is oftline applied on the prefilled matrix to group users. Users with similar interests will be clustered into the same group. The last step is to find nearest neighbors of the target user online and recommend items. In this paper, MovieLens is used as the test dataset. It is proved that the algorithm the article proposed can effectively alleviate data sparsity, improve poor real - time and the accuracy of recommendation to some extent.