Amazon Polarity
Dataset containing Amazon product reviews over 18 years, with positive or negative annotations. Each example includes a title, text content, and a polarity label. Great for training sentiment classification models.
Around 4 million reviews (train + test), JSON format with title, content and binary label fields
Apache 2.0
Description
Amazon Polarity is a massive corpus of English-speaking customer reviews, annotated for the classification of feelings into positive or negative. The data covers a period of 18 years and includes review titles and content.
What is this dataset for?
- Training sentiment classification models for e-commerce
- Improve customer feedback recommendation and analysis systems
- Test the robustness of NLP models on large quantities of real data
Can it be enriched or improved?
The dataset can be enriched by adding finer annotations (e.g. specific aspects of products), or merged with other review sources to create multilingual corpora.
🔎 In summary
🧠 Recommended for
- Data scientists
- NLP developers
- Sentiment analysis researchers
🔧 Compatible tools
- Hugging Face Transformers
- TensorFlow
- PyTorch
- Scikit-learn
💡 Tip
Combine this dataset with detailed reviews for more contextual and accurate learning.
Frequently Asked Questions
Does this dataset cover multiple languages?
No, reviews are mostly in English.
What labels are used for classification?
Two binary labels: 0 for negative (notes 1 and 2), 1 for positive (notes 4 and 5).
Can this dataset be used for tasks other than feeling classification?
Primarily designed for sentiment classification, but it can be exploited for general text analysis.
