Fake News Detection Using BERT Model with Joint Learning

Shishah, Wesam

doi:10.1007/s13369-021-05780-8

Cited by 22 publications

(14 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• An improved ablation study is implemented which shows that having the NERT is better than having the RFC in the proposed model. This paper is a continuation of our previous work [7]. The main difference from the previous work is that this paper proposes a similar model in different domains, specifically for Arabic fake news.…”

Section: Introductionmentioning

confidence: 89%

“…The proposed method presented in this paper is a continuation from our previous work [7]. In this paper we use similar components to the previous work [7]; therefore, in this section we summarise the main components.…”

Section: Proposed Methodologymentioning

confidence: 99%

“…Once BERT has been encoded using the primary encoder's features, the framework's hidden states can be fully utilise by the primary encoder's features. The main differences between this work and the previous study [7], is that this paper implements jointBERT in Arabic language, while the previous paper focuses on English. Even by using a similar method, working in another language is not trivial due to nature of the language and symbol differences.…”

Section: Introductionmentioning

confidence: 93%

“…The second challenge is that the Arabic language is very different from English language, thus a better text representation and text embedding is needed for the learning part of the detection algorithm. This paper proposes an implementation of the previous joint framework based on BERT [7]. The pretrained BERT uses a transformer encoder that helps to model the specific fake news data with trained general data.…”

Section: Introductionmentioning

confidence: 99%

“…The ablation study also aims to validate the parameter selection mechanism. Both experiments show the main motivation of component and parameter selection, which has not been done in the previous study [7]. This paper is organized as follows: A literature review of previous work on the problem is presented in Section II, followed by an in-depth explanation of the proposed methodology in Section III.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

JointBert for Detecting Arabic Fake News

Shishah

2022

IEEE Access

Self Cite

View full text Add to dashboard Cite

The rapid rise in the use of social media platforms has resulted in a recent surge of fake rumours, particularly among Arab countries. Such false information could potentially be detrimental to individuals and society. Detecting and blocking the spread of the fraudulent news in Arabic is critical. Many artificial intelligence algorithms, including contemporary transformer models, such as BERT, have been employed to detect the fake news in the past. Therefore, the fake news in Arabic can be detected using a revolutionary combined BERT architecture implemented in this paper. Extensive experiments were conducted to test the technique on real-world Arabic fake news datasets. In two of the fake news datasets, covid19fakes and Satirical, the suggested technique had a higher accuracy score than the current state-of-the-art Arabic fake news model. A comparable result can be achieved in other datasets; however, the proposed strategy fails to do so. All datasets except AraNews show an average F1 score improvement of 10% by implementing the proposed strategy. It was found that the proposed method was effective and superior to numerous other baselines of Arabic fake news detection models.

show abstract

Section: Introductionmentioning

confidence: 89%

Section: Proposed Methodologymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 93%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations