This is a python script for converting Stanford Sentiment Treebank dataset (https://nlp.stanford.edu/sentiment/index.html) to common used text classification format.
The stanfordSentimentTreebank.zip file is download from (http://nlp.stanford.edu/~socherr/stanfordSentimentTreebank.zip).
In the fine-grained sentiment classification dataset SST5, 0, 1, 2, 3, 4 represent very negative, negative, neutral, positive, very positive respectively.
In the binary sentiment classification dataset SST2, 0, 1 represent negative, positive respectively.
The statistics of dataset split for SST5 and SST2 are shown below.
| Dataset | SST5 | SST2 |
|---|---|---|
| Train | 8544 | 6920 |
| Dev | 1101 | 872 |
| Test | 2210 | 1821 |