Year: 2019
Type: conference-paper
Abstract: Authorship Verification (AV) is a type of stylometric analysis that addresses an authorship problem where, given a document of unknown origin and a set of documents written by a known author, the task is to identify whether the document is indeed written by that author.Previous research uses a number of techniques to address this problem.Most successful techniques in Classical Arabic as well as other languages use an SVM method supported by a distance measure in vector space and a distance/similarity threshold for accepting the document as authentic.While Arabic Authorship Attribution (where the task is to attribute the question document to one of several candidates) surveys and evaluates the usability of different distance measures, this paper is the first to provide such overview for Modern Arabic AV.Using a corpus of short texts from five common Modern Standard Arabic genres, this paper evaluates four common distance measures (Canberra, Manhattan, Cosine, and Jaccard) with a number of lexical, syntactic, and morphological features.The results show that Canberra Distance is a best performing distance measure in most genres, with an accuracy rate of up to 97.8%, well over highest known baseline.
Author: Hossam Ahmed
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2989978296