Amazon Polarity

The Amazon Polarity dataset is a set of reviews from Amazon. The dataset is constructed by taking review scores 1 and 2 as negative (class 1), and 4 and 5 as positive (class 2). Reviews with a score of 3 are ignored. The dataset spans a period of 18 years, including approximately 35 million reviews up to March 2013. Each class in the dataset has 1,800,000 training samples and 200,000 testing samples. The dataset includes product and user information, ratings, and a plaintext review.