DynaSent is an English-language benchmark task for ternary (positive/negative/neutral) sentiment analysis. DynaSent combines naturally occurring sentences with sentences created using the open-source Dynabench Platform, which facilities human-and-model-in-the-loop dataset creation. DynaSent has a total of 121,634 sentences, each validated by five crowdworkers.
Source: DynaSent: A Dynamic Benchmark for Sentiment Analysis
Papers
| Paper | Code | Results | Date | Stars |
|---|
