How to reproduce TACRED results

wang-research-lab / deepstruct

Code repo for ACL22 paper "DeepStruct: Pretraining of Language Models for Structure Prediction"

https://arxiv.org/pdf/2205.10475.pdf

Apache License 2.0

79 stars 4 forks source link

How to reproduce TACRED results #2

Open FlorianLR opened 1 year ago

FlorianLR commented 1 year ago

Hi,

Thanks for the paper and code!

I would like to use your model for Relation Extraction, starting with a reproduction of your 'multi-task w/ finetune' resuts on TACRED. However, I could not find a script running the experiment nor directions on how to do it.

Any help with this would be greatly appreciated!

Kind regards, Florian

Magolor commented 4 months ago

The TACRED dataset is already configured in src/data_processing/config.ini as well as in src/data_processing/dataset.py. You need to manually prepare the TACRED dataset as we don't own the copyright. The dataset should be prepared as json file aligning with the format in src/data_processing/dataset.py's TACRED class. The format is used the load_data_single_split.