Multiverse: Multilingual Evidence for Fake News Detection

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Information Science

Background:

• The proliferation of deceptive online information poses significant societal risks.
• Existing fake news detection technologies are often limited by their reliance on single-language data.
• Multilingual information is an underutilized resource in combating misinformation.

Purpose of the Study:

• To propose and evaluate a novel feature, Multiverse, for fake news detection.
• To demonstrate the efficacy of using cross-lingual evidence in identifying deceptive content.
• To improve the performance of existing fake news detection systems.

Main Methods:

• Development of the Multiverse feature, which incorporates multilingual evidence.
• Manual experiments to validate the hypothesis that cross-lingual evidence aids fake news detection.
• Comparison of a classification system utilizing Multiverse against baseline models on diverse datasets, including general news and COVID-19 misinformation.

Main Results:

• Manual experiments confirmed the utility of cross-lingual evidence for fake news detection.
• The Multiverse feature, when combined with linguistic features, significantly improved classification performance over baseline models.
• The proposed approach provides additional valuable signals for classifiers.

Conclusions:

• Multiverse represents a significant advancement in fake news detection technology.
• Incorporating multilingual evidence is crucial for developing more robust and accurate misinformation detection systems.
• The findings highlight the potential of cross-lingual analysis in addressing the global challenge of online deception.