Rotten Tomatoes: 10,662 Sentences for Binary Sentiment Classification

Encompassing 10,662 processed sentences from Rotten Tomatoes movie reviews, evenly split between 5,331 positive and 5,331 negative labels. Originally compiled by Bo Pang and Lillian Lee for a 2005 ACL publication, it serves as a foundational benchmark for binary sentiment classification. The data is provided by the Cornell Movie Review Data project and was last updated on Hugging Face in March 2024.

Use Cases

• Training binary sentiment classifiers using the 'text' and 'label' fields
• Benchmarking NLP model performance against historical 2005 ACL metrics
• Evaluating the impact of movie-specific vocabulary on text classification accuracy

Strengths

• Perfect class balance with exactly 5,331 records per category
• 10,662 total annotated records
• Peer-reviewed origin from the 2005 ACL 'Seeing stars' publication

Limitations

• Temporal staleness as the data was collected nearly two decades ago
• Sentence-level granularity lacks the context of full-length reviews
• Binary labels exclude nuanced or neutral sentiments

Provenance

Source

Bo Pang and Lillian Lee, Cornell University (ACL 2005)

Collection Method

Scraped and processed from Rotten Tomatoes movie reviews

Time Range

Pre-2005

Freshness

The source data is from 2005, meaning it lacks references to modern cinema, streaming platforms, or contemporary slang.

Geography

United States