Batura T.V. Techniques of determining author’s text style and their software implementation

Abstract: the article presents a review of formal methods of text attribution. The problem of determining the authorship of texts is present in different field and is important for philologists, literary critics, historians, lawyers. In solving the problem of text attribution the main interest and the main complexity is in the analysis of syntactic, lexical/idiomatic and stylistic levels of text. In a sense, a narrower task is in the text sentiment-analysis (defining the tone of the text). Techniques for solving the task can be useful for identifying authorship of the text. Unfortunately, expert analysis of author’s style is complex and time consuming. It’s desirable to find new approaches, allowing at least partially automate experts’ work. Therefore the article pays special attention exactly to the formal methods of author’s identification and software implementation of such methods. Currently, algorithms of data compression, methods of mathematical statistics, probability theory, neural networks algorithms and cluster analysis algorithms are applied for text attribution. The article describes the most popular software systems for author’s style identification for Russian language. Author attempts to make a comparative analysis, identify features and drawbacks of the reviews approaches. Among the problems hindering researches in text attribution there are a problem of selecting linguostylistic parameters of the text and a problem of selecting sample texts. The author states that there is a need in further researches, aimed at finding new or improving existing methods of texts attribution, at finding new characteristics allowing to clearly separate author’s style, including cases of short texts and small number of sample texts.

Keywords: