Author identification with feature transformation method

Abstract

Abstract Over the last few decades, there has been tremendous growth in online communication through different types of media. Communication via the Internet is anonymous, which causes a critical issue regarding identity tracing. Authorship identification can apply to tasks such as identifying an anonymous author, detecting plagiarism, or finding a ghostwriter. Previous research has outlined the various methods and their improvements for the identification of anonymous authors based on stylometry. However, changes in the writing style of an author over a long period has not been addressed. In this article, we propose a methodology for author identification where the writing style of an author changes. The proposed methodology consists of two phases: the first will show the change in writing style of the author and in another phase the change is mitigated by a new feature normalization technique. A novel Transform Feature to Current Time function is proposed for normalization, where features are shifted to current time and made available for further classification. A machine-learning algorithm is used to identify an author candidate. The experiments of the proposed methodology conducted on a set of text samples by several authors were collected over a different time period and the results show an improvement in performance.