Authorship verification for different languages, genres and topics

Abstract

Authorship verification is a branch of forensic authorship analysis addressing the following task: Given a number of sample documents of an author A and a document allegedly written by A , the task is to decide whether the author of the latter document is truly A or not. We present a scalable authorship verification method that copes with this problem across different languages, genres and topics. The central concept of our method is a model, which is trained with Dutch, English, Greek, Spanish and German text documents. The model sets for each language specific parameters and a threshold that accepts or rejects the alleged author as A . The proposed method offers a wide range of benefits, e.g., a universal (static) threshold for each language and scalability regarding almost any involved component (classification function, ensemble strategy, features, etc.). Furthermore, the method benefits from low runtime due to the fact that no natural language processing techniques nor other computationally-intensive methods are involved. In our experiments, we applied the method on 28 test corpora including 4525 verification cases across 16 genres and a huge number of mixed topics, where we achieved competitive results (75% median accuracy). With these results we were able to outperform two state-of-the-art baselines, given the same training and test corpora

Similarity Learning for Authorship Verification in Social Media

“…< τ1. On the other hand, for same-topic/different-author instances the output of Equation (11) shall be larger than a second predefined threshold τ2, i.e. d x…”

“…A token can refer to a word, abbreviation or punctuation symbol. As mentioned in [11], the number of known documents in each instance varies from one to ten. In [11] all documents in Dknown are concatenated into one single document, dknown.…”

Improving author verification based on topic modeling

“…2. Profile-based paradigm: All available samples in D known are first concatenated in one big document and a single representation vector is extracted (author's profile) (Ding et al, 2019;Halvani et al, 2016;Kocher & Savoy, 2017;Potha & Stamatatos, 2014). The methods in this paradigm follow an author-centric approach where the differences between documents by the same author are disregarded.…”

“…Intrinsic verification methods: They only analyze samples in D known and d u . Essentially, they consider author verification as a one-class classification task and attempt to estimate whether d u is similar enough to D known (Halvani et al, 2016;Jankowska et al, 2014;Potha & Stamatatos, 2014). Intrinsic models are usually robust since they do not depend on external resources.…”

Combining Textual and Speech Features in the NLI Task Using State-of-the-Art Machine Learning Techniques

“…NLI has a wide variety of potential applications and both its techniques and findings can be used in areas such as Second-Language Acquisition (Ortega, 2009), author profiling (Rangel et al, 2013), and authorship contribution (Halvani et al, 2016). Typically, NLI is employed as a starting point for investigations into crosslinguistic influence, see e.g.…”