Year: 2018
Type: dissertation
Abstract: This thesis explores a range of authorship attribution approaches and proposes new techniques to improve performance.Authorship attribution is the task of identifying the author of a text.It has attracted attention due to its relevance to a wide range of applications including forensic investigation and plagiarism detection.An array of features and approaches have been applied to this task.However, there has been a lack of study which involves multiple datasets or uses a range of different classifiers.Therefore, in this thesis we explore both neural and non-neural network models and use different feature representations on multiple datasets.We begin with a short introduction to authorship attribution in Chapter 1.A more comprehensive review of authorship attribution and its related tasks is given in Chapter 2. In Chapter 3 we introduces a novel analysis using topic modeling to examine the conditions under which each type of authorship attribution feature is useful.Chapter 4 explores the implementation of language modeling for authorship attribution.We describe the feature selection issue in standard authorship attribution approaches and evaluate whether n-gram language modeling can help to address the problem.Furthermore, we implement A Long Short Term Memory (LSTM) language model for authorship attribution and assess its effectiveness for the task.In Chapter 5 we present our work on using continuous representations for authorship attribution.In contrast to previous work, which uses discrete feature representations, our model learns continuous representations for n-gram features via a neural network jointly with the classification layer.The proposed model outperforms the state-of-the-art on two datasets, while producing comparable results on the remaining two.In addition, we describe our novel extension of the proposed models and show how the analysis in Chapter 3 helps to improve the attribution accuracy.Finally, we demonstrate how the authors' demographic profiles can help improve task performance via Multi Task Learning (MTL).In Chapter 6 we highlight the contributions of this thesis and propose directions for future research in this area.iiiFirst of all, I would like to thank my supervisors Mark Stevenson and Andreas Vlachos for their continuous support and guidance through my PhD.Mark and Andreas provided me with invaluable advice, help, and fruitful discussion.They shaped me as a researcher and made the PhD journey more enjoyable.I truly appreciate all the time and effort they took for me.I am very grateful to my PhD panel committee members, Lucia Specia and Gordon Fraser, for their feedback and constructive comments.My examiners
Author: Yunita Sari
Institution: University of Sheffield
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2901402029
