用户群组发现及兴趣用户推荐的改进的K-Means聚类算法

Improved K-Means Clustering Algorithm for User Group Discovery and Interest User Recommendation

摘要: 当前，电子商务网站、学习资源平台网站以及社交网站普遍都具备对评价事物的推荐功能，而不具备给用户发现和推荐感兴趣用户的功能。针对此问题，本文在K-means聚类算法的基础上加入包含抽样、降维、层次聚类等过程的预聚类阶段，设计出为用户推荐兴趣用户的FPSHK-means群组发现算法，并通过其与经典K-means聚类算法的对照实验，验证了FPSHK-means群组发现算法能比经典K-means算法发现更多的群组，且聚类结果更贴近数据对象的实际分布情况。

Abstract: At present, e-commerce websites, learning resource platforms websites and social networking sites generally have the recommendation function for comment things, but not the function of users discovering and recommending users who are interested. In this paper, the FPSHK-means group discovery algorithm was designed by blending a pre-clustering stage on the basis of the K-means clustering algorithm. The pre-clustering stage includes sampling, dimensionality reduction and hierarchical clustering. The FPSHK-means group discovery algorithm is designed to recommend interested users for the users. Through the comparison experiment with the classical K-means clustering algorithm, it is verified that the FPSHK-means group discovery algorithm can find more groups than the classical K-means algorithm. And the result of clustering is closer to the actual distribution of the data object.