Crowdsourced Text Classification Datasets from Crowdflower

Multiple text classification datasets sourced from Crowdflower, containing between 10,000 and 100,000 data points. The datasets are monolingual English and were created by crowdsourced annotation for tasks like sentiment classification and fact-checking.

Use Cases

• Train a sentiment classification model on monolingual English text data with crowdsourced labels.
• Fine-tune a fact-checking model using text data annotated by crowd workers.
• Analyze patterns in crowdsourced annotations for text classification tasks like sentiment analysis.

Strengths

• Contains between 10,000 and 100,000 data points across multiple datasets.
• Includes data for multiple text classification tasks such as sentiment classification and fact-checking.
• Datasets are structured for machine learning use with the mlcroissant library.

Limitations

• Specific column names, row counts per dataset, and file formats are unknown.
• Data is limited to monolingual English text, restricting multilingual applications.
• Crowdsourced annotations may introduce label noise or inconsistency.

Provenance

Source

tasksource via Hugging Face

Collection Method

Crowdsourced data collection via the Crowdflower platform.

Freshness

Last updated on 2023-06-21.

Geography

Primarily United States (Region:us).