A Modified Language Modeling Method for Authorship Attribution

This paper presents an approach to a closed-classauthorship attribution (AA) problem. It is based on languagemodeling for classification and called modified languagemodeling. Modified language modeling aims to offer a solutionfor AA problem by Combinations of both bigram wordsweighting and Unigram words weighting. It makes the relationbetween unseen text and training documents clearer with givingextra reward of training documents; training document includingbigram word as well as unigram words. Moreover, IDF valuemultiplied by related word probability has been used, instead ofremoving stop words which are provided by Stop words list. weevaluate Experimental results by four approaches; unigram,bigram, trigram and modified language modeling by using twoPersian poem corpora as WMPR-AA2016-A Dataset and WMPRAA2016-B Dataset. Results show that modified language modelingattributes authors better than other approaches. The result onWMPR-AA2016-B, which is bigger dataset, is much better thananother dataset for all approaches. This may indicate that ifadequate data is provided to train language modeling themodified language modeling can be a good solution to AAproblem.

Keywords:

Authorship Attribution , Authorship Identification , Language Modeling , Text Processing