Abstract: A novel semi-supervised learning algorithm fully exploring the inner semantic information to compensate for the limited message length is presented. The key idea is to explore term correlation data,which well captures the semantic information for term weighting and provides greater context for short texts. Direct and indirect dependency weights between terms are defined to reveal the semantic correlation between terms. Must-link and cannot-link are encoded as constraints for terms. This paper formulates microblog clustering problem as a semi-supervised non-negative matrix factorization co-clustering framework,which takes advantage of knowledge of features as pair-wise constraints. Extensive experiments are conducted on two real-world microblog datasets. Experimental results show that the effectiveness of the proposed algorithm. It not only greatly reduces the labor-intensive labeling process,but also deeply exploits the hidden information from microblog itself.

Key words: microblog, term correlation relationship, pair-wise constraints, semi-supervised clustering, non-negative matrix factorization

摘要： 针对微博文本内容短、稀疏、高维等特点,提出一种改进的半监督微博聚类算法。该算法利用词项间的关 系丰富文本特征,通过定义词项文档间关联关系和词项文档内关联关系揭示词项间语义的关联程度,并由此自动 生成有标记的数据来指导聚类过程。对词项先验信息进行成对约束编码,构建基于词项间成对约束的三重非负矩 阵分解模型来实现微博的半监督聚类。实验结果表明,该算法可以减少繁琐的人工标记过程,并能高效地进行微 博聚类。

关键词: 微博, 词项关联关系, 成对约束, 半监督聚类, 非负矩阵分解

CLC Number: 

• TP18