Implementation of Cross-Sentence N-ary Relation Extraction using Lower-Arity Universal Schemas

This repository contains implementations of the proposed method and baseline methods tested in our paper "Cross-Sentence N-ary Relation Extraction using Lower-Arity Universal Schemas" (To be appeared in EMNLP 2019).

Data

Following data files are required to run codes. Also, see data/README.md for the dataset format.

Wiki-90k and WF-20k dataset is available here.

data/glove.6B.300d.txt # You can download it from here (http://nlp.stanford.edu/data/glove.6B.zip).
data/Wiki-90k/train
data/Wiki-90k/dev
data/Wiki-90k/test
data/WF-20k/train.json
data/WF-20k/dev.json
data/WF-20k/test.json

Proposed method and baseline methods

Required environment

Codes of our proposed method are tested in the following environment.

• python (3.6.8)
  • numpy (1.16.2)
  • scikit-learn (0.20.3)
  • pytorch (1.1.0)
  • torch_scatter (1.3.1, https://github.com/rusty1s/pytorch_scatter)
  • networkx (2.2)

How to run

Before running codes, create logs directory. Results of experiments will be output in log files logs/ExpLog_<suffix>_<exp_number>.log. You can set suffix and exp_number by options when you start experiments.

To run experiments with the same settings in the paper, execute commands described in example.sh.

NOTES:

• You can use GPU by specifying its id by --gpu <id> option, or the codes will use CPU (slow).
• You can set a name and number of an experiment by using --suffix <NAME> and --exp_number <NUM> options.