jiaweisheng / cored

Source code for SIGIR 2022 paper: CorED: Incorporating Type-level and Instance-level Correlations for Fine-grained Event Detection.

CorED

Source code for SIGIR2022 paper: CorED: Incorporating Type-level and Instance-level Correlations for Fine-grained Event Detection.

Event detection (ED) is a pivotal task for information retrieval, which aims at identifying event triggers and classifying them into pre-defined event types. This paper simultaneously incorporates both the type-level and instance-level event correlations, and proposes a novel framework, termed as CorED. Specifically, we devise an adaptive graph-based type encoder to capture type-level correlations, learning type representations not only from their training data but also from their relevant types, thus leading to more informative type representations especially for the low-resource types. Besides, we devise an instance interactive decoder to capture instance-level correlations, which predicts event instance types conditioned on the contextual typed event instances, leveraging co-occurrence events as remarkable evidence in prediction. Empirical results demonstrate the unity of both type-level and instance-level correlations, and the model achieves effectiveness performance on both benchmarks.

Requirements

We conduct our experiments on the following environments:

python 3.6
CUDA: 9.0
GPU: Tesla T4
pytorch == 1.1.0
transformers == 4.9.1

Datasets

We adopt MAVEN and ACE-2005 as our datasets. The original MAVEN dataset can be accessed at this repo.

How to run

To run the code, you could run as following steps:

1. Put pretrained language models into ./plm/bert-base-uncased. Put original MAVEN dataset into ./dataset/maven/maven.

2. Run data preprocess as follows:

cd ./dataset/maven
python data_process.py

3. Conduct training/validation/testing as follows:

CUDA_VISIBLE_DEVICES=0 python -u main_cls.py --data_type maven --prefix exp_model  --do_train true --do_valid true --do_test true 

The hyper-parameters are recorded in ./utils/params.py.

Citation

If you find this code useful, please cite our work:

@inproceedings{Sheng2022:CorEE,
  title     = {CorED: Incorporating Type-level and Instance-level Correlations for Fine-grained Event Detection},
  author    = {Jiawei Sheng and Rui Sun and Shu Guo and Shiyao Cui and Jiangxia Cao and Lihong Wang and Tingwen Liu and Hongbo Xu},
  booktitle = {SIGIR},
  year      = {2022}
}

cored's Issues

type_literal_embeddings.npy

你好，请问"./dataset/maven/label_graphs/type_literal_embeddings.npy"这个文件是如何得来的呢？可否提供代码以供参考，谢谢！！！

如何让模型运行ACE005数据集