Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/littlepan0413/DuEE_baseline
事件抽取基线模型
https://github.com/littlepan0413/DuEE_baseline
Last synced: 7 days ago
JSON representation
事件抽取基线模型
- Host: GitHub
- URL: https://github.com/littlepan0413/DuEE_baseline
- Owner: littlepan0413
- License: apache-2.0
- Created: 2020-02-19T03:25:40.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2023-02-16T01:31:50.000Z (over 1 year ago)
- Last Synced: 2024-07-10T23:25:20.291Z (4 months ago)
- Language: Python
- Size: 1.37 MB
- Stars: 11
- Watchers: 1
- Forks: 1
- Open Issues: 3
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
English | [简体中文](./README.zh.md)
# Event Extraction Baseline model (EE-Baseline)
EE-Baseline is a event extraction baseline model for the event extraction dataset DuEE 1.0. This model splits event extraction into two sub-tasks: trigger extraction and argument extraction, solved as two sequence labeling problems in a pipelined fashion.
#### Trigger extraction model based on sequence labeling (Tri-SeqL)
Trigger extraction aims to predict whether a token triggers an event. We formulates trigger extraction as a sequence labeling problem with labels indicate the event types. This model is built upon a pre-trained language model ERNIE combined with a CRF layer.
> For above example, the model recognizes the trigger "求婚" and assigns it to the label "B-结婚" "I-结婚", finally, we get event type is "结婚"。
#### Argument extraction model based on Sequence labeling (Arg-SeqL)
Argument extraction aims to extract arguments and corresponding argument roles they play. We formulates argument extraction as a sequence labeling problem with labels indicate the argument roles. This model is also built upon a pre-trained language model ERNIE combined with a CRF layer.
> For above example, the model recognizes the arguments: 1) "李荣浩", assigns it labels "B-求婚者"、"I-求婚者" "I-求婚者"; 2) "杨丞琳", assigns it labels "B-求婚对象" "I-求婚对象" "I-求婚对象". Finally, we get argument roles and argument pairs is <求婚者, 李荣浩> <求婚对象, 杨丞琳>.
#### other versions
- [DuEE-PaddleHub](./DuEE-PaddleHub): Code simplified version base on PaddleHub
## Getting Started
### Environment Requirements
- python 2.7.x
- paddlepaddle-gpu >= 1.5.0 (see more details about PaddlePaddle [PaddlePaddle Homepage](https://www.paddlepaddle.org.cn/install/quick))
The code is tested on a single GPU Tesla K40m, with CUDA version=10.1, GPU Driver Version=418.39.
###### Install dependencies
```
pip install -r ./requirements.txt
```
### Integration steps
##### Step 1: Data preparation
Including steps Download pre-trained ERNIE model、Examples process and Schmea process
```shell
sh bin/script/data_preparation.sh
```
Re-executing this step requires deleting files `./model/ERNIE_1.0_max-len-512.tar.gz`、`./data/train.json`、`./data/dev.json`、`./data/test.json`、`./dict/vocab_trigger_label_map.txt`、`./dict/vocab_roles_label_map.txt`
##### Step 2: Train and results process
Including steps train and predict results of Tri-SeqL、train and predict results of Arg-SeqL、Prediction results process.
```shell
sh bin/script/train_and_eval.sh
```
Re-executing this step requires deleting files `./save_model/trigger`、`./save_model/role`、 `./save_model/trigger/pred_trigger.json` and `./save_model/role/pred_role.json`.
### Detailed steps
##### Step 1: Download pre-trained ERNIE model
```shell
cd ./model
wget https://ernie.bj.bcebos.com/ERNIE_1.0_max-len-512.tar.gz --no-check-certificate
mkdir ERNIE_1.0_max-len-512
tar -zxvf ERNIE_1.0_max-len-512.tar.gz -C ERNIE_1.0_max-len-512
```
Download ERNIE1.0 Base（max-len-512）model and extract it into `./model/ERNIE_1.0_max-len-512/`
##### Step 2: Examples process
Process examples into `./data/`, create files `train.json`、`dev.json` and `test.json`
```python
python bin/data_process.py origin_events_process ./data/eet_events.json ./data/
```
##### Step 3: Schmea process
- Trigger label process for Tri-SeqL and save file into `./dict/vocab_trigger_label_map.txt`
```python
python bin/data_process.py schema_event_type_process ./dict/event_schema.json ./dict/vocab_trigger_label_map.txt
```
- Argument Role Label for Arg-SeqL and save file into `./dict/vocab_roles_label_map.txt`
```python
python bin/data_process.py schema_role_process ./dict/event_schema.json ./dict/vocab_roles_label_map.txt
```
##### Step 4: Train Tri-SeqL
```shell
cd ./bin
HERE=$(readlink -f "$(dirname "$0")")
cd ${HERE}/..
DATA_DIR=${HERE}/../../data
PRETRAIN_MODEL=${HERE}/../../model/ERNIE_1.0_max-len-512
SAVE_MODEL=${HERE}/../../save_model
DICT=${HERE}/../../dict
GPUID=0
TRIGGER_SAVE_MODEL=${SAVE_MODEL}/trigger
sh script/train_event_trigger.sh ${GPUID} ${DATA_DIR} ${TRIGGER_SAVE_MODEL} ${PRETRAIN_MODEL} ${DICT}
```
##### Step 5: Prediction Tri-SeqL
```shell
cd ./bin
HERE=$(readlink -f "$(dirname "$0")")
cd ${HERE}/..
DATA_DIR=${HERE}/../../data
PRETRAIN_MODEL=${HERE}/../../model/ERNIE_1.0_max-len-512
SAVE_MODEL=${HERE}/../../save_model
DICT=${HERE}/../../dict
GPUID=0
TRIGGER_SAVE_MODEL=${SAVE_MODEL}/trigger
sh script/predict_event_trigger.sh ${GPUID} ${DATA_DIR} ${PRETRAIN_MODEL} ${TRIGGER_SAVE_MODEL}/final_model ${DICT}
```
##### Step 6: Train Arg-SeqL
```shell
cd ./bin
HERE=$(readlink -f "$(dirname "$0")")
cd ${HERE}/..
DATA_DIR=${HERE}/../../data
PRETRAIN_MODEL=${HERE}/../../model/ERNIE_1.0_max-len-512
SAVE_MODEL=${HERE}/../../save_model
DICT=${HERE}/../../dict
GPUID=0
ROLE_SAVE_MODEL=${SAVE_MODEL}/role
sh script/train_event_role.sh ${GPUID} ${DATA_DIR} ${ROLE_SAVE_MODEL} ${PRETRAIN_MODEL} ${DICT}
```
##### Step 7: Prediction Arg-SeqL
```shell
cd ./bin
HERE=$(readlink -f "$(dirname "$0")")
cd ${HERE}/..
DATA_DIR=${HERE}/../../data
PRETRAIN_MODEL=${HERE}/../../model/ERNIE_1.0_max-len-512
SAVE_MODEL=${HERE}/../../save_model
DICT=${HERE}/../../dict
GPUID=0
ROLE_SAVE_MODEL=${SAVE_MODEL}/role
sh script/predict_event_role.sh ${GPUID} ${DATA_DIR} ${PRETRAIN_MODEL} ${ROLE_SAVE_MODEL}/final_model ${DICT}
```
##### step 8: Prediction results process
- Transform test set（`./data/test.json`）to evaluation format file `./result/gold.json`
```python
python bin/predict_eval_process.py test_data_2_eval ./data/test.json ./result/gold.json
```
- Integrate and transform prediction results into evaluation format
Integrate predict results file of Tri-SeqL（`./save_model/trigger/pred_trigger.json`）、predict results file of Arg-SeqL（`./save_model/role/pred_role.json`）、events schema file（`./dict/event_schema.json`）into evaluation format file `./result/pred.json`
```python
python bin/predict_eval_process.py predict_data_2_eval ./save_model/trigger/pred_trigger.json ./save_model/role/pred_role.json ./dict/event_schema.json ./result/pred.json
```
# Evaluation
Zip your prediction json (`./result/pred.json`) file and submit it to official website
## Discussion
If you have any question, you can submit an issue in github and we will respond periodically.