opinion_mining_with_syn_cons

This repositry contains our code, configurations, and model for our work on "A Unified Span-Based Approach for Opinion Mining with Syntactic Constituents", which is published on NAACL-2021. The src directory contains our code and the exp-4.1-baseline contains our experiment for "Baseline+BERT" (data0, the first data of the five fold cross-validation).

Environment

Python3, Pytorch, Transformers 2.1.1 (for BERT)

Data

MPQA2.0 url PTB and OntoNotes can be download from LDC.

Training

Please reset and check the files in the train.sh and config.json when you want to run the code.

sh train.sh GPU\_ID

Test

To test the performance of the trained model, you should run the following script.

sh predict.sh GPU\_ID

We release the sample model of the "exp-4.1-baseline" on the Google Drive, url. Important, use the offline evaluation script to eval the output file.