SST-2 Sentiment Analysis
Dataset
The binary classification version of the Stanford Sentiment Treebank, with 67,000 movie review sentences, is one of the core tasks for sentiment analysis and the GLUE benchmark suite.
Dataset Highlights
The most classic sentiment analysis benchmark in the NLP field, widely used for model evaluation and transfer learning research
Real Movie Review Data
Data sourced from the Rotten Tomatoes movie review website, split into single sentences by the Stanford Parser, truly reflecting user sentiment expression patterns.
Binary Classification Simplified Design
Original fine-grained 5-level sentiment annotations simplified to positive/negative binary classification, reducing task complexity and facilitating quick validation of model capabilities.
Core Tasks of the GLUE Benchmark
As one of the official tasks in the GLUE benchmark suite, SST-2 is a standard option for evaluating the performance of pre-trained language models like BERT and GPT.
Transfer Learning Friendly
With a moderate scale of 67,000 labeled samples, it is very suitable for fine-tuning pre-trained models and is a commonly used evaluation set for few-shot and zero-shot learning.
Rich Academic Literature
Published by Socher et al. in 2013 at EMNLP, it has been cited tens of thousands of times, with comprehensive documentation, papers, and reproduction code.
Stanford NLP Production
Originating from the Stanford University Natural Language Processing Group, the data quality is strictly controlled, making it a gold standard dataset for NLP research and education.
Applicable Scenarios
From model evaluation to industrial applications, covering various needs for sentiment analysis
Sentiment Analysis
Train and evaluate text sentiment classification models to determine the positive or negative tendency of comments, feedback, and other texts
Text Classification
As a general text classification benchmark, validate the classification performance of architectures such as CNN, RNN, and Transformer
Transfer Learning
Used to fine-tune pre-trained models like BERT, RoBERTa, and DeBERTa, quickly validating the effectiveness of downstream tasks
GLUE Benchmark Evaluation
As a core sub-task of the GLUE benchmark suite, used for multi-task learning and comparison of language model capabilities
Data Preview
The following is a JSON format example of the SST-2 dataset, containing positive and negative sentiment samples
[
{
"sentence": "a stirring , funny and finally transporting re-imagining of beauty and the beast and 1930s horror films",
"label": 1
},
{
"sentence": "apparently reassembled from the cutting-room floor of any given daytime soap",
"label": 0
},
{
"sentence": "they presume their audience wo n't sit still for a contemplative cool-eli thriller",
"label": 0
},
{
"sentence": "the greatest musicians playing the greatest music ever written",
"label": 1
},
{
"sentence": "a beautiful , fun and often affecting movie",
"label": 1
},
{
"sentence": "if you sometimes like to go to the movies to have fun , wasabi is a good place to start",
"label": 1
},
{
"sentence": "emerges as something rare , an intelligent thriller that 's both moving and suspenseful",
"label": 1
},
{
"sentence": "unflinchingly bleak and desperate",
"label": 0
}
]
# label: 1 = positive (positive), 0 = negative (negative)
3 Steps to Get Started Quickly
From browsing to analysis, you can start your NLP project in just a few minutes
Browse Dataset
View dataset details on the Ace Data Cloud platform, understand field descriptions, sample size, and data distribution metadata.
Download Data
Obtain the complete dataset containing 67,349 labeled sentences, covering training, validation, and test set divisions.
Load and Train
Use datasets.load_dataset("sst2") to load the data, quickly fine-tune a pre-trained model or train a custom classifier.
Start Exploring SST-2 Sentiment Analysis Data
A classic NLP benchmark dataset with open research licenses, available immediately. Whether you are a deep learning researcher or an NLP engineer, SST-2 is the preferred dataset for sentiment analysis tasks.