GNN-based Short text similarity algorithm

Requirements

pickle

collections

functools

torchtext

numpy

pytorch

tqdm

scipy

dgl

sklearn

prettytable

Dataset

STS-B:

STS-B contains 8,628 sentence pairs, of which 5,749 are training data, 1,379 are test data, and 1,500 are development data. STS-B can get from http://alt.qcri.org/semeval2017/index.php?id=tasks

SICK:

There are 4439 pairs of training data, 4906 pairs of test data, and 495 pairs of trial data in the SICK data set. SICK can be gotten from http://alt.qcri.org/semeval2014/task1/.

Usage

1. Set parameters in fig.py
2. Run test.py

The experimental results will be saved in log.txt The model is saved as model.bt.