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
Variants: DynaSent
This dataset is used in 1 benchmark:
Recent papers with results on this dataset: