Masking topic‐related information to enhance authorship attribution

Stamatatos, Efstathios

doi:10.1002/asi.23968

Cited by 33 publications

(39 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This bias helps representations of vocabulary-based models to be close. However, such features fail to identify authors in cross-domain scenarios (Stamatatos, 2018), while our model focuses less on topic-and semantic-related words but achieves comparable performance (even better for SimRank and the authorship attribution task).…”

Section: In-depth Analysismentioning

confidence: 91%

Representation learning of writing style

Hay

Doan

Popineau³

et al. 2020

Proceedings of the Sixth Workshop on Noisy User-Generated Text (W-Nut 2020)

View full text Add to dashboard Cite

In this paper, we introduce a new method of representation learning that aims to embed documents in a stylometric space. Previous studies in the field of authorship analysis focused on feature engineering techniques in order to represent document styles and to enhance model performance in specific tasks. Instead, we directly embed documents in a stylometric space by relying on a reference set of authors and the intra-author consistency property which is one of two components in our definition of writing style. The main intuition of this paper is that we can define a general stylometric space from a set of reference authors such that, in this space, the coordinates of different documents will be close when the documents are by the same author, and spread away when they are by different authors, even for documents by authors who are not in the set of reference authors. The method we propose allows for the clustering of documents based on stylistic clues reflecting the authorship of documents. For the empirical validation of the method, we train a deep neural network model to predict authors of a large reference dataset consisting of news and blog articles. Albeit the learning process is supervised, it does not require a dedicated labeling of the data but it relies only on the metadata of the articles which are available in huge amounts. We evaluate the model on multiple datasets, on both the authorship clustering and the authorship attribution tasks.

show abstract

Section: In-depth Analysismentioning

confidence: 91%

Representation learning of writing style

Hay

Doan

Popineau³

et al. 2020

Proceedings of the Sixth Workshop on Noisy User-Generated Text (W-Nut 2020)

View full text Add to dashboard Cite

show abstract

“…Recently, cross-genre, cross-domain, cross-topic and multi-topic data sets [14,20] have been studied in AA tasks. Most studies on topic and domain issues have focused on cross-domain or cross-topic tasks [21–23]. Topic influence has been addressed through generative models where function words and stylometric markers have been used in joint inference of the author and the topic [22].…”

Section: Related Workmentioning

confidence: 99%

“…Topic influence has been addressed through generative models where function words and stylometric markers have been used in joint inference of the author and the topic [22]. On the other hand, studies on removing topic-related information through masking document content has been successfully applied in multi-topic data sets [21]. In another study, a syntactic feature set has been suggested for topic-independent AA [23].…”

Section: Related Workmentioning

confidence: 99%

Bucketed common vector scaling for authorship attribution in heterogeneous web collections: A scaling approach for authorship attribution

Agun

Yılmazel

2019

Journal of Information Science

View full text Add to dashboard Cite

Domain, genre and topic influences on author style adversely affect the performance of authorship attribution (AA) in multi-genre and multi-domain data sets. Although recent approaches to AA tasks focus on suggesting new feature sets and sampling techniques to improve the robustness of a classification system, they do not incorporate domain-specific properties to reduce the negative impact of irrelevant features on AA. This study presents a novel scaling approach, namely, bucketed common vector scaling, to efficiently reduce negative domain influence without reducing the dimensionality of existing features; therefore, this approach is easily transferable and applicable in a classification system. Classification performances on English-language competition data sets consisting of emails and articles and Turkish-language web documents consisting of blogs, articles and tweets indicate that our approach is very competitive to top-performing approaches in English competition data sets and is significantly improving the top classification performance in mixed-domain experiments on blogs, articles and tweets.

show abstract

“…In particular, style of documents can be used to infer their genre or reveal information about their authors (Stamatatos, 2009). Authorship analysis, dealing with the personal style of authors, is a very active research area (Neal et al, 2018;Rocha et al, 2017;Seroussi, Zukerman, & Bohnert, 2014;Stamatatos, 2018).…”

Section: Introductionmentioning

confidence: 99%

Improving author verification based on topic modeling

Potha

Stamatatos

2019

Asso for Info Science & Tech

Self Cite

View full text Add to dashboard Cite

Authorship analysis attempts to reveal information about authors of digital documents enabling applications in digital humanities, text forensics, and cyber‐security. Author verification is a fundamental task where, given a set of texts written by a certain author, we should decide whether another text is also by that author. In this article we systematically study the usefulness of topic modeling in author verification. We examine several author verification methods that cover the main paradigms, namely, intrinsic (attempt to solve a one‐class classification task) and extrinsic (attempt to solve a binary classification task) methods as well as profile‐based (all documents of known authorship are treated cumulatively) and instance‐based (each document of known authorship is treated separately) approaches combined with well‐known topic modeling methods such as Latent Semantic Indexing (LSI) and Latent Dirichlet Allocation (LDA). We use benchmark data sets and demonstrate that LDA is better combined with extrinsic methods, while the most effective intrinsic method is based on LSI. Moreover, topic modeling seems to be particularly effective for profile‐based approaches and the performance is enhanced when latent topics are extracted by an enriched set of documents. The comparison to state‐of‐the‐art methods demonstrates the great potential of the approaches presented in this study. It is also demonstrates that even when genre‐agnostic external documents are used, the proposed extrinsic models are very competitive.

show abstract

Masking topic‐related information to enhance authorship attribution

Cited by 33 publications

References 38 publications

Representation learning of writing style

Representation learning of writing style

Bucketed common vector scaling for authorship attribution in heterogeneous web collections: A scaling approach for authorship attribution

Improving author verification based on topic modeling

Contact Info

Product

Resources

About