Stanford Sentiment Treebank is a standard sentiment dataset, mainly used for sentiment classification, in which each node of the sentence analysis tree has a fine-grained sentiment annotation.

The dataset was released by the NLP group at Stanford University, which contains 239,232 sentences and phrases. Compared with most sentiment prediction systems that ignore word order, this deep learning model builds a complete representation based on sentence structure. It can judge sentiment based on phrases composed of words.

This dataset was released by the Natural Language Processing Group of Stanford University in 2013. The related paper is "Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank".