README

Code for TriMF: "A Trigger-Sense Memory Flow Framework for Joint Entity and Relation Extraction". accepted at WWW 2021. For details of the model and experiments, please see our paper.

Setup

Requirements

pip install -r requirements.txt

Datasets

We use the DyGIE and SpERT scripts to pre-process the datasets, see and follow their README: DyGIE and SpERT.

Dataset format:

[
    {
        "tokens": ["allan", "chernoff", "live", "from", "the", "new", "york", "stock", "exchange", "with", "more", "."], 
        "entities": [{"type": "PER", "start": 0, "end": 2}, {"type": "FAC", "start": 5, "end": 9}], 
        "relations": [{"type": "PHYS", "head": 0, "tail": 1}], "orig_id": "CNN_ENG_20030530_130025.12-4", 
        "dependency": [{"tail": 0, "head": 1, "type": "nsubj"}, {"tail": 1, "head": 1, "type": "ROOT"}, {"tail": 2, "head": 1, "type": "advmod"}, {"tail": 3, "head": 2, "type": "prep"}, {"tail": 4, "head": 8, "type": "det"}, {"tail": 5, "head": 8, "type": "amod"}, {"tail": 6, "head": 7, "type": "compound"}, {"tail": 7, "head": 8, "type": "compound"}, {"tail": 8, "head": 3, "type": "pobj"}, {"tail": 9, "head": 1, "type": "prep"}, {"tail": 10, "head": 9, "type": "pobj"}, {"tail": 11, "head": 1, "type": "punct"}], 
        "ltokens": ["aol", "time", "warner", "and", "microsoft", "are", "burying", "the", "hatchet", "."], 
        "rtokens": ["bring", "us", "up", "to", "speed", "."]
    }
]

Examples

Train:

After pre-processing the data, save the datasets under data, and run:

python trimf.py train --config configs/example.conf

Evaluate:

Download checkpoints from this link, and save in the data folder, and run:

python trimf.py eval --config configs/batch_eval.conf

Citation

If you have any questions related to the code or the paper, feel free to email syl@zju.edu.cn.

@inproceedings{10.1145/3442381.3449895,
author = {Shen, Yongliang and Ma, Xinyin and Tang, Yechun and Lu, Weiming},
title = {A Trigger-Sense Memory Flow Framework for Joint Entity and Relation Extraction},
year = {2021},
isbn = {9781450383127},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3442381.3449895},
doi = {10.1145/3442381.3449895},
booktitle = {Proceedings of the Web Conference 2021},
pages = {1704–1715},
numpages = {12},
location = {Ljubljana, Slovenia},
series = {WWW '21}
}

Acknowledgement

DyGIE from https://github.com/luanyi/DyGIE

SpERT from https://github.com/markus-eberts/spert.git

SciBERT from https://github.com/allenai/scibert

tricktreat / trimf