Unsupervised Multi-Author Document Decomposition Based on Hidden Markov Model

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 706–714, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Unsupervised Multi-Author Document Decomposition Based on Hidden Markov Model Khaled Aldebei Xiangjian He Wenjing Jia Global Big Data Technologies Centre University of Technology Sydney Australia {Khaled.Aldebei,Xiangjian.He,Wenjing.Jia}@uts.edu.au Jie Yang Lab of Pattern Analysis and Machine Intelligence Shanghai Jiaotong University China Jieyang@sjtu.edu.cn Abstract This paper proposes an unsupervised approach for segmenting a multi- author document into authorial com- ponents. The key novelty is that we utilize the sequential patterns hid- den among document elements when determining their authorships. For this purpose, we adopt Hidden Markov Model (HMM) and construct a sequen- tial probabilistic model to capture the dependencies of sequential sentences and their authorships. An unsuper- vised learning method is developed to initialize the HMM parameters. Exper- imental results on benchmark datasets have demonstrated the significant ben- efit of our idea and our approach has outperformed the state-of-the-arts on all tests. As an example of its applica- tions, the proposed approach is applied for attributing authorship of a docu- ment and has also shown promising re- sults. 1 Introduction Authorship analysis is a process of inspect- ing documents in order to extract autho- rial information about these documents. is considered as a general concept that em-It braces several types of authorship subjects, in- cluding authorship verification, plagiarism detection and author attribution. Author- ship verification (Brocardo et al., 2013; Potha and Stamatatos, 2014) decides whether a given document is written by a specific author. Pla- giarism detection (Stein et al., 2011; Keste- mont et al., 2011) seeks to expose the simi- larity between two texts. However, it is un- able to determine if they are written by the same author. In author attribution (Juola, 2006; Savoy, 2015), a real author of an anony- mous document is predicted using labeled doc- uments of a set of candidate authors. Another significant subject in author- ship analysis, which has received compara- tively less attention from research commu- nity, is authorship-based document decompo- sition (ABDD). This subject is to group the sentences of a multi-author document to dif- ferent classes, of which each contains the sen- tences written by only one author. Many ap- plications can take advantage of such a sub- ject, especially those in forensic investigation, which aim to determine the authorship of sen- tences in a multi-author document. Further- more, this kind of subject is beneficial for de- tecting plagiarism in a document and defining contributions of authors in a multi-author doc- ument for commercial purpose. ABDD can also be applied to identify which source (re- garded as an ‘author’ in this paper) a part of a document is copied from when the doc- ument is formed by taking contents from var- ious sources. In despite of the benefits of ABDD, there has been little research reported on this sub- ject. Koppel et al. (2011) are the first re- searchers who implemented an unsupervised approach for ABDD. However, their approach …

Content truncated to 500 words