# Multi-View-of-Expert-for-Chinese-Relation-Extraction

**Repository Path**: tmg-nudt/multi-view-of-expert-for-chinese-relation-extraction

## Basic Information

- **Project Name**: Multi-View-of-Expert-for-Chinese-Relation-Extraction
- **Description**: PAKDD23: <Dynamic Multi-View Fusion Mechanism For Chinese Relation Extraction>

- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2023-03-07
- **Last Updated**: 2024-12-04

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Introduction
code for < PAKDD2023 > < Dynamic Multi-View Fusion Mechanism For Chinese Relation Extraction>


# Requirements
General

- Python (verified on 3.8)
- CUDA (verified on 11.1/10.2)
- torch (1.7.1)

```
pip install -r requirments.txt
```
# Quick Start
### dataset

```
/workspace/MoVE_V0/data (31.89MB)                                                                                                                                                 
├── SanWen (1.66MB)                                                                                                                                                               
│   ├── train.txt (1.35MB)                                                                                                                                                        
│   ├── rel2id.txt (103b)                                                                                                                                                         
│   ├── test.txt (168.93KB)                                                                                                                                                       
│   └── valid.txt (145.42KB)                                                                                                                                                      
├── Toy (11.96KB)                                                                                                                                                                 
│   ├── toy.txt (11.91KB)                                                                                                                                                         
│   └── rel2id.txt (46b)                                                                                                                                                          
├── FinRE_withNA (3.32MB)                                                                                                                                                         
│   ├── train.txt (2.39MB)                                                                                                                                                        
│   ├── rel2id.txt (527b)                                                                                                                                                         
│   ├── test.txt (682.01KB)                                                                                                                                                       
│   └── valid.txt (272.12KB)                                                                                                                                                      
├── FinRE (2.41MB)                                                                                                                                                                
│   ├── train.txt (1.73MB)                                                                                                                                                        
│   ├── rel2id.txt (484b)                                                                                                                                                         
│   ├── test.txt (494.76KB)                                                                                                                                                       
│   └── valid.txt (201.78KB)                                                                                                                                                      
├── eda.py (1.12KB)                                                                                                                                                               
├── SciRE (6.15MB)                                                                                                                                                                
│   ├── train.txt (4.3MB)                                                                                                                                                         
│   ├── dev.txt (660.95KB)                                                                                                                                                        
│   ├── rel2id.txt (46b)                                                                                                                                                          
│   └── test.txt (1.2MB)                                                                                                                                                          
├── clearn.py (2.56KB)                                                                                                                                                            
├── SciRE_withNA (15.87MB)                                                                                                                                                        
│   ├── train.txt (11.22MB)                                                                                                                                                       
│   ├── dev.txt (1.57MB)                                                                                                                                                          
│   ├── rel2id.txt (46b)                                                                                                                                                          
│   └── test.txt (3.09MB)                                                                                                                                                         
└── SanWen_withNA (2.47MB)                                                                                                                                                        
    ├── train.txt (1.99MB)                                                                                                                                                        
    ├── rel2id.txt (102b)                                                                                                                                                         
    ├── test.txt (265.03KB)                                                                                                                                                       
    └── valid.txt (221.64KB)  
```

### train
```
bash shells/run_finre.sh
```

```
bash shells/run_sanwen.sh
```

```
bash shells/run_scire.sh
```




### hyper-param [model/config_layer.py]
```
# ======================

config_encoder = dict(
    use_radical=True,
    use_lexicon=True,
    use_synoym=False,
    multiview_feature_dim=100,
    hidden_dim=512,
    pooler_dim=100,
)
### TODO


## 这里的input_dim和hidden_dim 随着 特征的维度会变化
config_fuse = dict(
    input_dim=1368,  # [300, 768， 868, 968]
    hidden_dim=1368,  # [300, 768， 868, 968]
    num_layer=1,
)


config_classifier = dict(
    input_dim=1368,  # [300, 768, 868, 968]
    hidden_dim=512,
    pooling_type="logsumexp",  # ['sum', 'logsumexp', 'max', 'min', 'CLS']
)

config_semantic = dict(
    use_char=True,  #  CharCNN
    use_biword=True,
    char_embed_size=100,
    biword_embed_size=100,
    pooler_dim=100,
    use_bert=True,
    bert_path="./hf_models/chinese-bert-wwm-ext",
    # bert_path="./hf_models/chinese-lert-base",
    # bert_path="./hf_models/chinese-macbert-large",
)

config_lexicon = dict(
    fix_embedding=False,  # 是否对于外部词典的embedding进行finetune
    use_char=True,
    use_biword=True,
    use_count=True,
    word_emb_dim=50,
    biword_emb_dim=50,
    gaz_embed_dim=50,
    pooler_dim=100,
)


config_radical = dict(
    radical_feature_dim=100,
    pooler_dim=100,
    filter_nums=[50, 30, 20],
    kernel_sozes=[5, 3, 2],
    pool_method="max",
)


# ======================

CONFIG = dict(
    config_lexicon=config_lexicon,
    config_sematic=config_semantic,
    config_radical=config_radical,
    config_encoder=config_encoder,
    config_fuse=config_fuse,
```

# Citation

# License
The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License for Noncommercial use only. Any commercial use should get formal permission first.