A better indicator for genre classification: Topic word or surface text feature: A case study of recognition of brief biography

Classification based on topic (content) rather than genre (form) prevails in the text data mining and search engine circle. To simplify this work, a BOW (Bag of Words) strategy, counting topic-related words as features, is comprehensively utilized to make a final decision. Indeed, texts can be categorized by expression styles rather than their themes. Brief biography is a typical text class which differs from others in its genre. In order to differentiate biographical data from other texts quickly and effectively, we constructed two knowledge bases and compared their performances, i.e. topic keyword (content) and simple linguistic features (form). The results showed that the latter form-based approach is superior to its former counterpart. A further discussion is elaborated at the end of this paper.