Abstract: Topic detection achieves quite good result in the traditional media research. This paper discussed the refiness and performance evaluation of the topic detection technique in the new kind of medics such as microblog, proposed theM13-SinglePass topic detection algorithm on the basis of the structured information such as the relationships of attenlions and fans between contacts, the inner connection relationships such as forwarding and comment between posts. Beside considering the above microblog characteristics,MB-SinglePass introduces the characteristics extension technique in order w enrich characteristics information. At the same time, the paper used the combined similarity aiming at the shortage of singly utilizing the Jaccard similarity cocfficient,cosine based similarity and semantic similarity. Compared with the traditional algorithms,MI3-SinglePass shows better performance on the actual dataset of sing microblog. Additionally, experiment according to the similarity strategy reveals better result by using combined similarity than singular similariy.

Key words: Microblog,SinglePass,Topic detection,Text similarity,Synonyms extension