MIT Movies Named Entity Recognition Dataset with Error Corrections

Presenting a corrected version of the MIT movies dataset for named entity recognition, created to showcase the Galileo Data Intelligence Platform. It contains crowdsourced annotations for token classification tasks, with errors in the original dataset identified and 4% of samples fixed.

Use Cases

• Train a named entity recognition model on crowdsourced token classification annotations.
• Analyze the impact of error correction on model performance using the fixed 4% of samples.
• Benchmark NER systems on a monolingual English dataset with crowdsourced labels.

Strengths

• Dataset includes corrections for identified annotation errors and ill-formed samples.
• Fixes applied to 4% of the original dataset based on error pattern analysis.
• Designed for the specific task of named entity recognition (NER).

Limitations

• Original dataset size, row count, and column structure are unknown.
• Limited to monolingual English data, which may not generalize to other languages.
• Crowdsourced annotations may introduce label noise or inconsistency.

Provenance

Source

Original MIT movies dataset, modified by rungalileo.

Collection Method

Crowdsourced annotations, with errors identified and corrected using the Galileo platform.

Freshness

Last updated on 2022-10-25.

Geography

Region is specified as 'us'.