ExplainabilityTransparencyAccountabilitycs.CL
A New cross-domain strategy based XAI models for fake news detection
Deepak Kanneganti
February 4, 2023
arXiv.org
2 citations
Key Takeaways
-
Explainability is essential for trustworthy fake news detection: The paper emphasizes that understanding why a model flags content as fake is crucial, not just whether it does so accurately. This transparency helps users and regulators trust automated fact-checking systems.
-
Different explanation methods work better at different stages: The research tested multiple explainability tools (Anchor, LIME, SHAP, and ELI5) and found that different approaches are optimal depending on the level of analysis, suggesting that no single explanation method fits all situations in fake news detection.
-
Models can transfer knowledge across different types of content: By using a "cross-domain strategy," the researchers showed that a model trained on fake news in one domain (e.g., political news) can be adapted to detect misinformation in other domains (e.g., health or finance), making fake news detection systems more versatile and efficient.
-
Pre-trained language models provide a strong foundation: The use of fine-tuned BERT demonstrates that starting with large, general-purpose language models and adapting them for specific tasks is an effective approach for building more reliable fake news detection systems.
Abstract
In this study, we presented a four-level cross-domain strategy for fake news detection on pre-trained models. Cross-domain text classification is a task of a model adopting a target domain by using the knowledge of the source domain. Explainability is crucial in understanding the behaviour of these complex models. A fine-tune BERT model is used to. perform cross-domain classification with several experiments using datasets from different domains. Explanatory models like Anchor, ELI5, LIME and SHAP are used to design a novel explainable approach to cross-domain levels. The experimental analysis has given an ideal pair of XAI models on different levels of cross-domain.