# sprel

**Repository Path**: xinyi_chen/sprel

## Basic Information

- **Project Name**: sprel
- **Description**: code and data for paper: Relation as Text: A Semantic-Preserving Method for Relational Triple Extraction
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-08-20
- **Last Updated**: 2024-08-20

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

Code Structure:
-> SPRel Code
--> data
----> data_raw: original dataset without our preprocess. 
----> rel2id.json : all relations with their text and id.
----> data_preprocess.py : our preprocess method.
----> parser.py : SpaCy parser for preprocess
----> dataset_new.pickle : preprocessed dataset. its about 1GB! cannot upload for review
----> dataset_small.pickle : samples of preprocessed dataset. each train, valid, and test dataset for nyt, nyt_star, webnlg, webnlg_star has only 10 samples for review. use python to review them.
----> SPRelDataset.py : dataset loading module for training and testing.
----> utils.py
----> sample.txt : a single sample for review.
--> model
----> Metrics.py
----> SPRelModels.py
--> data_statistics.ipynb
--> nyt_star_epe.ipynb : training of SPRel-EPE on nyt_star
--> nyt_star_re.ipynb : training of SPRel-RE on nyt_star
--> nyt_star_joint_eval.ipynb : joint evaluation for SPRel on nyt_star
--> webnlg_star_epe.ipynb
--> webnlg_star_re.ipynb
--> webnlg_star_joint_eval.ipynb
--> README
Suggested Reading:
1. check data/sample.txt. this is a complete preprocessed sample.
2. (optional) check data/data_preprocess.py and data_statistics.py if you want to understand how we preprocess data.
3. (optional) check data/SPRelDataset.py. this is how we load data for training.
4. check model/SPRelModels.py. this file modifies BERT model from transformers. modifications can be easily searched by keyword ''#modified''. our models are ''BertForEntityPairExtraction'' and ''BertForRelationExtraction'' at the bottom of the file.
5. (optional) check model/Metric.py.
6. check nyt/webnlg_star_epe/re/joint_eval.ipynb in the main folder. our training and evalution details are in them
Special Notice:
1. The code is only a demonstration for review with some unnecessary files deleted.