Unsupervised Multi-Author Document Decomposition Based on Hidden Markov Model

Abstract

This paper proposes an unsupervised approach for segmenting a multiauthor document into authorial components.The key novelty is that we utilize the sequential patterns hidden among document elements when determining their authorships. For this purpose, we adopt Hidden Markov Model (HMM) and construct a sequential probabilistic model to capture the dependencies of sequential sentences and their authorships. An unsupervised learning method is developed to initialize the HMM parameters. Experimental results on benchmark datasets have demonstrated the significant benefit of our idea and our approach has outperformed the state-of-the-arts on all tests. As an example of its applications, the proposed approach is applied for attributing authorship of a document and has also shown promising results.

SUDMAD: Sequential and unsupervised decomposition of a multi‐author document based on a hidden markov model

“…Furthermore, we randomly select four articles on the authorship‐based document decomposition topic (i.e., the same topic addressed in this article). The articles are Koppel et al (), Giannella (), Daks and Clark (), and Aldebei, He, Jia, and Yang (). The lengths of these articles are 257, 215, 104, and 229 sentences, respectively.…”

“…We consider the contextual information hidden among a series of sentences and propose to use the Hidden Markov Model (HMM) to explore the sequential patterns in the document. Note that the initial idea of this work has recently been published in the ACL conference (Aldebei, He, Jia, & Yang, ). A simple HMM was constructed to find a useful sequential correlation between consecutive sentences of the document, which achieved very encouraging results.…”

“…In this paper, we further extend our work and propose a two‐stage HMM model to utilize the sequential patterns among sentences more comprehensively. Apart from more details and experiments that are included to disclose the benefit of this idea, this paper distinguishes from our previous work (Aldebei et al, ) significantly in the following three new contributions: We propose to utilize the useful sequential correlations among the consecutive sentences to determine the authorial components and construct a two‐stage HMM called “SequentialUD” — Sequential Unsupervised Decomposition, to model the relationships between authorships and sentences. To further boost the performance of this approach, a boosted HMM learning procedure is proposed. The initial classification results obtained using the statistically learned and preliminary HMM are used to create a labeled training dataset to learn a more accurate HMM. Moreover, the contextual relationships among sentences are further utilized to refine the classification results and a refined version of the SequentialUD is proposed. …”

“…In this paper, we further extend our work and propose a two-stage HMM model to utilize the sequential patterns among sentences more comprehensively. Apart from more details and experiments that are included to disclose the benefit of this idea, this paper distinguishes from our previous work (Aldebei et al, 2016) significantly in the following three new contributions:…”

Sequential and Unsupervised Document Authorial Clustering Based on Hidden Markov Model

“…In [8], the authors presented an unsupervised approach for the same problem, which utilized the difference of the posterior probabilities of a Naive-Bayesian Model in order to improve the performance of the approach. Another approach has been presented in [9] for multi-author document segmentation. In this approach, a simple HMM was constructed in order to segment the sentences of a multi-author document into authorial components.…”