newsmediabias/Bias-Question-Answering

Dataset Card for QA Bias Detection Dataset

Summary

• Description: This dataset is designed for the task of bias detection in text, particularly focusing on dimensions of ageism and sentiment analysis. It contains question-answer pairs that assess potential biases in statements.
• Purpose: To facilitate research and development in the areas of bias detection, natural language understanding, and sentiment analysis.
• Supported Tasks: Bias detection, sentiment analysis, natural language understanding, question answering.
• Languages: English

Composition

• Size of Dataset: 3,900 rows.
• Column Names: ['text', 'dimension', 'aspect', 'biased_words', 'sentiment', 'aggregate_label', 'Bias Type', 'Question', 'Answer', one column for indexing].
• Data Format: Each record contains a text statement, its associated dimensions and aspects of bias, biased words identified, sentiment, aggregate label for bias, bias type, and a QA pair for bias detection.

Source Data

• Initial Data Collection and Normalization: curated

Annotations

• Annotation process: Annotations include bias type, sentiment, and question-answer pairs evaluating the presence and type of bias in the text.
• Who are the annotators?: Human annotation.

Personal and Sensitive Information

• Considerations: The dataset contains text that may include sensitive topics related to bias and sentiment.

Considerations for Using the Data

• Social Impact of Dataset: The dataset could be valuable for developing systems that detect and mitigate bias in textual data.
• Other Known Limitations: The dataset's effectiveness is limited to the scope of its annotations and the quality of its text sources.