Scholarlink

<p>The necessity of online marketplaces is under serious jeopardy due to the proliferation of false reviews. Current detection approaches commonly rely on large amounts of labeled data, computationally costly adversarial settings, and have restricted interpretability, all of which make them unable to generalize across domains. Hence, this paper proposes CROSS-FRD, a self-supervised transformer framework for detecting fraudulent reviews across domains, with the aim of resolving these deficiencies. To capture linguistic cues that are domain-agnostic, our framework uses Masked Review approaching (MRM). With the use of Contrastive Domain Calibration (CDC), our method can be able to align reviews that share semantic similarities across categories. Together, we can learn representations efficiently from unlabeled data with little computational overhead; that is, our goal. The results on three benchmark datasets Yelp, Amazon, and TripAdvisor demonstrated the efficiency of our proposed approach, CROSS-FRD outperformed the state-of-the-art methods such as BERT-FT, DANN-BERT, and CORAL-BERT, increasing cross-domain F1-scores by up to 12.3% and AUC by 0.10%. To provide visibility into the outcomes of our proposed approach and demonstrate the trust and transparency in the obtained results, we also offer a comprehensive explanation for our approach results using Shapely Additive Explanations (SHAP) method and attention techniques. The experimental results prove that our proposed framework CROSS-FRD is an effective, practical, scalable, and interpretable strategy for cross-domain fraud detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A self-supervised transformer framework for cross-domain fraudulent review detection (CROSS-FRD)

  • 期刊论文

  • 发表日期：

    2025年11月19日

查看全文
  • Ali Al-yousef,
  • Rami Mohawesh,
  • Ahmad A. Saifan,
  • Moy’awiah A. Al-Shannaq

摘要

The necessity of online marketplaces is under serious jeopardy due to the proliferation of false reviews. Current detection approaches commonly rely on large amounts of labeled data, computationally costly adversarial settings, and have restricted interpretability, all of which make them unable to generalize across domains. Hence, this paper proposes CROSS-FRD, a self-supervised transformer framework for detecting fraudulent reviews across domains, with the aim of resolving these deficiencies. To capture linguistic cues that are domain-agnostic, our framework uses Masked Review approaching (MRM). With the use of Contrastive Domain Calibration (CDC), our method can be able to align reviews that share semantic similarities across categories. Together, we can learn representations efficiently from unlabeled data with little computational overhead; that is, our goal. The results on three benchmark datasets Yelp, Amazon, and TripAdvisor demonstrated the efficiency of our proposed approach, CROSS-FRD outperformed the state-of-the-art methods such as BERT-FT, DANN-BERT, and CORAL-BERT, increasing cross-domain F1-scores by up to 12.3% and AUC by 0.10%. To provide visibility into the outcomes of our proposed approach and demonstrate the trust and transparency in the obtained results, we also offer a comprehensive explanation for our approach results using Shapely Additive Explanations (SHAP) method and attention techniques. The experimental results prove that our proposed framework CROSS-FRD is an effective, practical, scalable, and interpretable strategy for cross-domain fraud detection.