yelp數(shù)據(jù)集最新數(shù)據(jù)資料，適合NLP分類任務(wù)，情感分類等。In 2015, Yelp held a contest in which it asked participants to predict the rating of
a restaurant given its review. Zhang, Zhao, and Lecun (2015) simplified the dataset by converting the 1- and 2-star ratings into a “negative” sentiment class and the 3- and 4-star ratings into a “positive” sentiment class, and split it into 560,000 training samples and 38,000 testing samples.