A Profile-Based Authorship Attribution Approach to Forensic Identification in Chinese Online Messages

With the popularity of Internet technologies and applications, inappropriate or illegal online messages have become a problem for the society. The goal of authorship attribution for anonymous online messages is to identify the authorship from a group of potential suspects for investigation identification. Most previous contributions focused on extracting various writing-style features and employing machine learning algorithms to identify the author. However, as far as Chinese online messages are concerned, they contain not only Chinese characters but also English characters, special symbols, emoticons, slang, etc. It is challenging for word segmentation techniques to segment Chinese online messages correctly. Moreover, online messages are usually short. The performance for short samples would be decreased greatly using traditional machine learning algorithms. In this paper, a profile-based authorship attribution approach for Chinese online messages is firstly provided. N-gram techniques are employed to extract frequency sequences, and the category frequency feature selection method is used to filter common frequent sequences. The profile-based method is used to represent the suspects as category profiles. The illegal messages are attributed to the most likely authorship by comparing the similarity between unknown illegal online messages and suspects’ profiles. Experiments on BBS, Blog, and E-mail datasets show that the proposed profile-based authorship attribution approach can identify the authors effectively. Compared with two instance-based benchmark methods, the proposed profile-based method can obtain better authorship attribution results.