Cross-domain deception detection using support vector networks

Hernández-Castañeda, Ángel; Calvo, Hiram; Gelbukh, Alexander; Flores, Jorge J. García

doi:10.1007/s00500-016-2409-2

Cited by 38 publications

(22 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…However, our best performance is achieved by fine-tuning BERT embeddings (Devlin et al, 2018) on this task. On the hotel review dataset, bootstrap validation accuracy is 90.5%, just behind the 91.2% reported by Feng et al (2012) who combine bag-of-words with constituency tree fragments.…”

Section: Introductionmentioning

confidence: 74%

Fact or Factitious? Contextualized Opinion Spam Detection

Kennedy¹,

Walsh²,

Sloka³

et al. 2019

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop

View full text Add to dashboard Cite

In recent years, it has been shown that falsification of online reviews can have a substantial, quantifiable effect on the success of the subject. This creates a large enticement for sellers to participate in review deception to boost their own success, or hinder the competition. Most current efforts to detect review deception are based on supervised classifiers trained on syntactic and lexical patterns. However, recent neural approaches to classification have been shown to match or outperform state-of-the-art methods. In this paper, we perform an analytic comparison of these methods, and introduce our own results. By fine-tuning Google's recently published transformer-based architecture, BERT, on the fake review detection task, we demonstrate near state-of-the-art performance, achieving over 90% accuracy on a widely used deception detection dataset.

show abstract

Section: Introductionmentioning

confidence: 74%

Fact or Factitious? Contextualized Opinion Spam Detection

Kennedy¹,

Walsh²,

Sloka³

et al. 2019

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop

View full text Add to dashboard Cite

show abstract

“…The first three examples in Table are derived directly from Table . While Hernandez et al () studied Amazon reviews, they did not mention the base rate, so this was derived from another source. Finally, the 0.5 base rate is used to represent a balanced sample, as well as that of the commonly studied crime suspect scenario (Fienberg et al, ).…”

Section: Determining Usefulnessmentioning

confidence: 99%

“…Researchers have explored ADDs that incorporate many different techniques in low base rate contexts. Tools using linguistic analysis appear most frequently in our review (Fuller et al, ; Hernández‐Castañeda et al, ; Larcker & Zakolyukina, ; Siering et al, ; Throckmorton et al, ; Zhang et al, ). That there are six studies that used linguistic analytics illustrates how far this research has come in the last 15 years.…”

Section: Introductionmentioning

confidence: 99%

“…The papers reviewed here demonstrate a wide variety in the reporting of performance measures. While some papers report simply overall accuracy (Cook et al, ; Hernández‐Castañeda et al, ; Honts & Amato, ), others only report ROC and/or AUC (Hobson et al, ; Throckmorton et al, ; Twyman et al, ). Several report multiple traditional SDT measures such as accuracy, precision, sensitivity, specificity, and/or false positive and negative (Farwell et al, ; Fuller et al, ; Twyman et al, ; Zhang et al, ).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Advancing the assessment of automated deception detection systems: Incorporating base rate and cost into system evaluation

Twitchell

Fuller

2018

Information Systems Journal

View full text Add to dashboard Cite

In the last two decades, there has been an increased interest in automated deception detection systems (ADDs) for use in screening, although little attention has been paid to the usefulness of these systems. ADDs use various means, both invasive and non-invasive, to ascertain individual intent to deceive or engage in malicious behaviour. Many papers introducing ADDs use signal detection theory to compare a technique's ability to detect malicious intent with other techniques, but in doing so, they do not include contextual information such as base rate and cost. In this paper, we aim to improve future research by showing how the inclusion of contextual information provides a more realistic picture of the research. Through both theoretical arguments and a real-data example, we show that especially for those contexts where malicious intent is infrequent (ie, with low base rates of deception) that not factoring in the base rate overestimates the accuracy and therefore usefulness. We conclude with recommendations for how future research should provide a fuller picture of the accuracy and usefulness of ADDs.

show abstract

“…Neste modelo um texto é representado como uma "sacola" de palavras, ou seja, guardamos apenas o número de ocorrência das palavras não levando em conta a ordem das palavras. Muitos autores HERNÁNDEZ-CASTAÑEDA et al, 2017; utilizaram o bag of words e obtendo resultados significativos em seus experimentos apesar da simplicidade do modelo.…”

Section: Bag Of Words (Bow)unclassified

Análise e caracterização de textos intencionalmente enganosos escritos em português usando métodos de processamento de textos

Okano¹

View full text Add to dashboard Cite

Atualmente a web é um ambiente em que pessoas postam e buscam informações sobre os mais diversos tópicos. No entanto, nem sempre uma informação encontrada na web é verdadeira já que alguns usuários maliciosos usam a web para disseminar informações falsas com o intuito de manipular ou enganar pessoas. Uma das maneiras de detectar estas informações falsas é através de processamento de textos. Atualmente há trabalhos direcionados à língua inglesa para identificação de textos enganosos, mas poucos trabalhos são voltados para a língua portuguesa. Neste trabalho, inicialmente foi criado um corpus paralelo de reviews de livros enganosos e foi feito um trabalho inicial de classificação automático do mesmo. Foi feito um estudo utilizando as research questions propostas por Hauch et al. para fazer uma análise psicolinguística do corpus de notícias falsas Fake.Br com o intuito de verificar as características mais relevantes para a classificação de fake news. Ainda utilizando o corpus Fake.Br treinamos algoritmos de aprendizagem de máquina supervisionados para fazer a classificação automática de fake news e utilizamos também um algoritmo de deep learning chamado Hierarchical attention network para verificar sua performance na detecção de fake news.

show abstract

Cross-domain deception detection using support vector networks

Cited by 38 publications

References 35 publications

Fact or Factitious? Contextualized Opinion Spam Detection

Fact or Factitious? Contextualized Opinion Spam Detection

Advancing the assessment of automated deception detection systems: Incorporating base rate and cost into system evaluation

Análise e caracterização de textos intencionalmente enganosos escritos em português usando métodos de processamento de textos

Contact Info

Product

Resources

About