摘要
通过对传统k-means算法优缺点的研究分析,提出一种改进的k-means聚类算法。随机初始化k/2个簇心,划分最大的簇并删除空簇,在更新簇心的同时判断簇心位置的合理性;及时对簇心做出修改,使得最后聚类出的k个簇中不会出现空簇;使用高斯核函数作为测量向量之间距离的方法,提高聚类的准确性。基于此改进的k-means算法,使用在不同网站上采集的文章作为数据源,并利用TF-IDF以及Word2Vec技术对文本进行向量化处理,进而完成对文本的聚类任务。与传统的k-means文本聚类相比,不仅提高了聚类的准确性,而且改善了传统k-means算法结果可能会出现空簇的缺陷。
Through the research and analysis of the advantages and disadvantages of the traditional k-means algorithm, we proposed an improved k-means clustering algorithm. We randomly initialized k /2 cluster cores, and divided the largest cluster and deleted the empty clusters. The cluster core was updated to determine the rationality of the cluster center position. The cluster core was modified in time to make the empty clusters would not appear in the last k clusters. The Gaussian kernel function was used as the method to measure the distance between vectors, which greatly improved the accuracy of clustering. Based on this improved k-means algorithm, articles collected on different websites were used as data sources, and we used TF-IDF and Word2Vec technologies to preprocess the text, and completed the task of clustering text. Compared with traditional k-means text clustering, it not only improves the accuracy, but also corrects the defect of empty clusters in the results of traditional k-means algorithm.
作者
张国锋
吴国文
Zhang Guofeng;Wu Guowen(College of Computer Science and Technology, Donghua University, Shanghai 200050, China)
出处
《计算机应用与软件》
北大核心
2019年第9期281-284,301,共5页
Computer Applications and Software