复现CNN 和 lstm+att 做关系抽取

##实验结果

cnn

Micro-averaged result (excluding Other): P = 2004/2262 = 84.36% R = 2004/2449 = 79.23.% F1 = 81.18%

MACRO-averaged result (excluding Other): P = 83.66% R = 77.15% F1 = 80.72%

a-cnn

此结果运行代码 https://github.com/lawlietAi/pytorch-acnn-model 所得

Micro-averaged result (excluding Other): P = 2004/2262 = 85.59% R = 2004/2449 = 79.83% F1 = 82.78%

MACRO-averaged result (excluding Other): P = 85.66% R = 78.15% F1 = 81.72%

lstm+att

Micro-averaged result (excluding Other): P = 2004/2262 = 83.59% R = 2004/2449 = 79.83% F1 = 81.75%

MACRO-averaged result (excluding Other): P = 82.66% R = 79.15% F1 = 80.72%

Usage

Train

#! /bin/bash

mkdir -p saved_models

CUDA_VISIBLE_DEVICES=2 python3  src/train.py  --num_epochs=200 --word_dim=50
python src/train.py  --num_epochs=200 --word_dim=50 --test
src/scorer.pl data/results.txt data/test_keys.txt 

python src/train.py  --num_epochs=200 --word_dim=300
python src/train.py  --num_epochs=200 --word_dim=300 --test

Data

SemEval-2010 Task #8

Reference

• Relation Classification via Convolutional Deep Neural Network (COLING 2014), D Zeng et al.
• Relation Extraction: Perspective from Convolutional Neural Networks (NAACL 2015), TH Nguyen et al.
• Zhou. (2016). Attention-Based Bidirectional Long Short-Term Memory Networks for Relation Classification. ACL
• https://github.com/lawlietAi/pytorch-acnn-model
• https://github.com/FrankWork/conv_relation