CAJ | 학술논문

针对传统的微博聚类分析中,只单独针对微博阅读数、评论数等数据（下称微博结构化数据）进行分类或者单独针对由微博内容进行文本分词得到的分词数据（下称微博分词）进行分类的问题,本文采用了Kohonen聚类,研究结合微博结构化数据和微博分词的融合数据聚类的效果是否比单独对微博结构化数据或对微博分词聚类有所提高。实证数据实验结果显示,微博结构化数据单独聚类会出现一个类的标准差特别大（本文称为离群类）,而对融合数据聚类,微博结构化数据则不会出现离群类;融合数据聚类结果对微博分词的影响不显著。
침대전통적미박취류분석중,지단독침대미박열독수、평론수등수거（하칭미박결구화수거）진행분류혹자단독침대유미박내용진행문본분사득도적분사수거（하칭미박분사）진행분류적문제,본문채용료Kohonen취류,연구결합미박결구화수거화미박분사적융합수거취류적효과시부비단독대미박결구화수거혹대미박분사취류유소제고。실증수거실험결과현시,미박결구화수거단독취류회출현일개류적표준차특별대（본문칭위리군류）,이대융합수거취류,미박결구화수거칙불회출현리군류;융합수거취류결과대미박분사적영향불현저。
This paper focused the problem that traditional clustering analysis have focused on only structured data such as microblog reading numbers and microblog comment numbers （rnicroblog segmentation） or only microblog text. In this paper, nil- croblog metadata are combined with microblog text to form fusion data and Kohonen Network Clustering is applied to test if fusion data clustering is better than microblog metadata clustering and than microblog text clustering. Experiments indicates that microblog metadata clustering may cause a class with large standard deviation （outlier class） and on the contrary, fusion data clustering does not. Microblog text clustering performs as well as fusion clustering.