Fake News Text Dataset for Model Training

Text data for training models to identify fake news. The dataset was created by the somosnlp-hackathon-2022 team and uploaded to Hugging Face in March 2022. The exact volume of articles and specific sources are not detailed in the provided metadata.

Use Cases

• Train a binary classifier to predict the 'label' field (e.g., real/fake) using text content features.
• Perform feature engineering on article text to extract linguistic patterns associated with misinformation.
• Fine-tune transformer models like BERT on the text data for sequence classification tasks.
• Analyze the dataset for potential biases in the 'label' distribution across different topics or sources.

Strengths

• Dataset is hosted on a reputable platform (Hugging Face) with version control.
• Platform tags indicate the dataset is in a tabular CSV format suitable for standard ML workflows.
• Includes text modality, which is core for NLP tasks like fake news detection.

Limitations

• No information on the number of rows, columns, or specific text features limits reproducibility assessment.
• The absence of a described license creates uncertainty for commercial use.
• Data is from 2022 and may not reflect current misinformation trends or social media language.

Provenance

Source

somosnlp-hackathon-2022 team via Hugging Face.

Collection Method

Presumably collected and labeled for a hackathon project; exact methodology unknown.

Time Range

null

Freshness

Last updated in March 2022; no update frequency specified.

Geography

Platform tag 'Regionus' suggests a possible US focus, but not confirmed.