Authorship identification is a task to identify the authorship of an anonymous text. It first analyzes given the text and the tries to find the authorship from authorship candidates. In order to analyze the text well, most of previous studies have developed useful linguistic features. Especially, they only focused on formal texts such as novel and journal.
Recently, people have written articles to express their opinions on the web. Therefore, the authorship identification for web text is also emerged. However, the previous studies are not directly applied to this task because the linguistic features only designed for the formal text are not suitable for the text written on web board.
This paper proposes various features for identifying authorship of an anonymous text written on the web board. Unlike the formal text, the text on the web board is influenced by author life patterns. For example, one person mainly writes a text at day time, on the other hand, the other person mostly writes a text at Sunday night. That is, the written time and day of week are important features to identify the web text authorship. In addition, web text contains lots of socio-linguistic characters such as emoticons and character repetitions. Since they reflect the style of author writing, they should be considered when we analyze the web text. Therefore we add the emoticons and character repetitions features. Finally, the proposed method combines not only these features but also the lexical features such as length of tokens and topic distribution that are useful to identify the authorship. According to the experimental results on a real data set, the proposed features are proved to be effective in identifying the authors. Furthermore, the method with the combination of the all features shows the best performance.