Misinformation Detection.pdf

A study introduced a misinformation verification system for Arabic COVID-19 related news on Twitter, using the ArCOV19-Rumors dataset. The dataset contains 3,584 annotated tweets, with 48.9% false and 51.1% true tweets. The system combines machine learning classification algorithms with semantic analysis, achieving 93% overall accuracy in detecting misinformation.

Methodology

1. Data collection: ArCOV19-Rumors dataset was used, containing 3,584 Arabic tweets.
2. Data preprocessing: Multiple phases of preprocessing techniques were applied.
3. Machine learning classification: Different algorithms were used, combined with semantic analysis.

Results

1. The model achieved 93% best overall accuracy in detecting misinformation.
2. A new dataset of COVID-19 related claims in Arabic was built to test the model, achieving 92% best accuracy.

Key Findings

1. The combination of machine learning techniques and linguistic analysis achieves high accuracy in detecting misinformation.
2. The system can help fact-checking services and journalists in immediately notifying false claims.
3. The study addresses the lack of research on misinformation detection in Arabic online content, particularly related to the COVID-19 pandemic.

Conclusion

The study provides a model for detecting misleading news related to the COVID-19 pandemic in Arabic language, using machine learning tools combined with semantic analysis. The results show high accuracy in detecting misinformation, making it a useful tool for fact-checking services and journalists.