Cross-domain Authorship Attribution: Author Identification using a Multi-Aspect Ensemble Approach - (CLEF 2019)

M Rahgouy, HB Giglou, T Rahgouy, MK Sheykhlan, E Mohammadzadeh

Author Attribution (AA) as one of the most important tasks of authorship analysis attracted huge body of research in recent years. In this task, given a document, the goal is to identify its author from a set of known authors and samples of their writings. In PAN 2019 shared tasks, the AA task is expanded in two ways. First, by having documents written by authors other than the known authors (UNK documents). Second, using a cross-domain set of documents. The task baseline and previous works mainly rely on character-level representation of documents because of their better generalization capability across different languages and domains. However, we hypothesize that ignoring coarse-grain features of documents may lead to loss of valuable information about the author's style. In this paper we propose an ensemble approach that combines models built upon different levels of document representation in order to investigate this hypothesis. Experimental results presented in this paper show that the coarse-grained representations of documents play an important role in identifying the authors style alongside the fine-grained representations.

Author Profiling: Bots and Gender Prediction using a Multi-Aspect Ensemble Approach - (CLEF 2019)

HB Giglou, M Rahgouy, T Rahgooy, MK Sheykhlan, E Mohammadzadeh

Author Profiling is one of the most important tasks in authorship analysis. In PAN 2019 shared tasks, the gender identification of the author is the main focus. Compared to the previous year the author profiling task is expended by having documents written by bots. In order to tackle this new challenge we propose a two phase approach. In the first phase we exploit the TF-IDF features of the documents to train a model that learns to detect documents generated by bots. Next, we train three models on character-level and word-level representations of the documents and aggregate their results using majority voting. Finally, we empirically show the effectiveness of our proposed approach on the PAN 2019 development dataset for author profiling.

Author Masking Directed by Author’s Style - (CLEF 2018)

M Rahgouy, HB Giglou, T Rahgooy, H Zeynali, SKM Rasooli

Author verification algorithms mainly rely on learning statistical fingerprints of authors. In the other hand, most of the previous algorithms in author masking try to apply changes to the original texts blindly without considering finger-prints and uses them to apply directed transformations and distortions to the original text.We represent author finger-prints with different statistics such as word choice distribution, sentence length preference, etc. obtained from author’s known texts. Automatic and manual evaluations of the obfuscated texts are very promising and show the effectiveness of our approach.