README.md

JRRELP

Knowledge Graph Enhanced Relation Extraction
George Stoica, Emmanouil Antonios Platanios, and Barnabás Póczos
NeurIPS 2020 KR2ML Workshop

This repository contains all code relevant to JRRELP, a general multitask learning framework for improving relation extraction via link prediction.

For details on this work please check out our:

• arXiv: Paper
• NeurIPS 2020 KR2ML Workshop: Paper & Poster

Running Experiments

We base our code off of three open-source repositories: PA-LSTM, C-GCN, & SpanBERT. To run JRRELP in a model, navigate to its respective named directory. We will use PALSTM as a running example.

1. Index into model directory
2. Download & prepare data by following instructions in the observed README.md file.
3. Run experiment by:

CUDA_VISIBLE_DEVICES=0 python train.py

3. To change parameters of an experiment, all you need to do is modify either "base_config.yaml" or "kglp_config.yaml" in the configs directory. Each config file has parameter comments highlighting what each parameter does.