Multilingual Text Corpus for Fake News Classification

Multilingual text data is provided for the task of fake news classification. The dataset contains news articles labeled as real or fake across multiple languages. The original author, creation date, and exact size are not specified in the provided information.

Use Cases

• Train a binary classifier to predict a 'fake_news_label' from the article 'text' content.
• Develop a multilingual model using the 'language' feature to detect misinformation across different regions.
• Analyze linguistic patterns and 'text' features that distinguish real from fake articles within each language group.
• Benchmark the performance of transformer models like BERT or XLM-RoBERTa on this cross-lingual classification task.

Strengths

• Dataset supports multilingual analysis, a key feature for global misinformation research.
• Core task of binary classification (real vs. fake) is clearly defined for model training.

Limitations

• The exact number of rows, articles per language, and class balance are unknown, making reproducibility difficult.
• Specific source credibility, annotation methodology, and potential label noise are not documented.
• Without a 'timestamp' column, the dataset's temporal relevance and coverage cannot be assessed.