SPPGAT

This repository is the source code for the paper 'Semantics from Look and Sound: Learning Chinese Character Representations from Function-Specific Components and Pronunciations'.

Steps to run the code

1. Build env with requirements.txt;
2. Download the BERT model 'chinese_L-12_H-768_A-12' and change the path in config file;
3. Download the data from 'Weibo' and change the path in config file;
4. Run:

bash script/run_weibo.sh