Authorship Attribution By Consensus Among Multiple Features

Proceedings of the 27th International Conference on Computational Linguistics, pages 2766–2777 Santa Fe, New Mexico, USA, August 20-26, 2018. 2766 Authorship Attribution By Consensus Among Multiple Features Jagadeesh Patchala University of Cincinnati Cincinnati OH, USA jagadeesh.patchala@gmail.com Raj Bhatnagar University of Cincinnati Cincinnati OH, USA raj.bhatnagar@uc.edu Abstract Most existing research on authorship attribution uses various lexical, syntactic and semantic fea- tures. In this paper we demonstrate an effective template-based approach for combining various syntactic features of a document for authorship analysis. The parse-tree based features that we propose are independent of the topic of a document and reflect the innate writing styles of authors. We show that the use of templates including sub-trees of parse trees in conjunction with other syntactic features result in improved author attribution rates. Another contribution is the demon- stration that Dempster’s rule based combination of evidence from syntactic features performs better than other evidence-combination methods. We also demonstrate that our methodology works well for the case where actual author is not included in the candidate author set. 1 Introduction In the past decade authorship analysis has become a significant need due to its various application areas such as identification of authors of anonymous posts in blogs (Koppel et al., 2011; Koppel and Yeron, 2014), of emails (Patchala et al., 2015), for copyright issues, plagiarism detection, and authentication of electronic documents (Abbasi and Chen, 2005; Chaski, 2005). Typically we encounter two types of author attribution problems. The first kind is the closed author set problem where the test-document is written by someone within the set of candidate authors being examined. The second kind is the semi- closed authorship problem where the test-document may or may not be written by someone in the set of candidate authors. The main hypothesis underlying authorship attribution methods is that the grammatical style of an author remains the same across topics and is fairly unique to each author. This uniqueness of an author is easily identifiable when considering small sets of authors, and it expectedly dilutes as the author set becomes large. It is, therefore, desirable to discover from texts more such features that are unique to an author and help increase the discriminability of authors based on their writings. The focus of this paper is on identification and integration of features to enable improved authorship attribution. We use template based classification and an evidence combination method for ex- ploiting information embedded in different types of syntactic features of documents. Figure 1: Height two, height three and height four sub-trees of a parse tree. We build templates using the frequencies of sub-trees of var- ious heights derived from the parse trees of sentences in doc- uments. Our intuitive basis for using these new features is that they capture the grammati- cal style very well. We consider the frequencies of sub-trees of heights two, three, and four, as This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/. 2767 possible features to describe an author’s writings. Examples …

Content truncated to 500 words