From rumor to genetic mutation detection with explanations: a GAN approach

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Computational Biology

Background:

• Social media facilitates rapid information spread, leading to a rise in misinformation and rumors.
• Traditional rumor detection methods relying on manual feature selection are insufficient for the scale of online content.
• Explainability is crucial for AI systems to ensure user trust in decision-making processes.

Purpose of the Study:

• To develop an explainable artificial intelligence (AI) model for text-based rumor detection using generative adversarial networks (GANs).
• To demonstrate the model's versatility by applying it to a gene classification with mutation detection task.
• To improve the accuracy and trustworthiness of misinformation detection systems.

Main Methods:

• A GAN-based layered model was proposed, incorporating both generative and discriminative components.
• The model was trained to detect rumors using only tweet-level texts, without external verified databases.
• Layered generators introduced controversial elements into non-rumors to train layered discriminators for identifying problematic text segments.

Main Results:

• The proposed model achieved superior performance in rumor detection on the PHEME dataset, outperforming state-of-the-art baselines by [Formula: see text] in macro-f1 score.
• The model demonstrated excellent performance in a gene mutation detection case study, achieving a macro-f1 score of [Formula: see text].
• The layered architecture enhanced the model's ability to identify specific problematic elements within text.

Conclusions:

• The GAN-based layered model offers an effective and explainable solution for text-based rumor detection.
• The approach is adaptable to various text classification problems, including biological data analysis.
• This method enhances AI trustworthiness by providing transparent decision-making for misinformation identification.