N-gram Based Text Classification According To Authorship
Anđelka Zečević · 2011
Authorship attribution studies consider author's identification of an anonymous text. This is a long history problem with a great number of various approaches. Those ones based on n-grams single out by their performances and good results. A n-gram approach is language independent but the selection of a number n is actually not. The focus of this paper is determination of a set of optimal values for number n for specific task of classification of newspaper articles written in Serbian according to authorship. We combine two different algorithms: the first one is based on counting common n-grams and the another one is based on relative frequency of n-grams. Experimental results are obtained for pairs of n-gram and profile sizes and it can be concluded that for all profile sizes the best results are obtained for 3≤n≤7. A goal of this paper is to identify authors of anonymous articles from the local daily newspapers using n-gram based algorithm. The articles discuss similar topics, all are written in Serbian and published in the same period of time. The scheme of our algorithm is depicted in Figure 1 and represents a classical profile-based algorithm: 1