Skip to content

Entry for the 19th "Citi Cup" Financial Innovation Application Competition 2024. Our submission, Fink_NER_ED_Web, is an extension of the Fink project for the 19th "Citi Cup" Financial Innovation Application Competition in 2024. Centered around the BERT model, it integrates two downstream tasks: entity extraction and disambiguation, as well as senti

Notifications You must be signed in to change notification settings

JerrySiRi/Fink_NER_ED_BondSenti_Web

 
 

Repository files navigation

BondSenti:基于BERT的债券违约情绪分析

0. 项目介绍

第19届“花旗杯”金融创新应用大赛参赛作品,获得全国总决赛二等奖。以BERT模型为核心,组合实体抽取和消歧、情绪分析两个下游任务,抽取出非结构化文本中的债券实体和公司实体,并分别对相应实体进行对应文本的情绪分析,为债券违约提供参考,并将模型封装部署到了Web端。

1. 运行环境

本项目使用python == 3.7.4开发,主要使用到的库版本如下,详情见requirements.txt

pytorch == 1.7.1 
pytorch-crf == 0.7.2  
pytorch-transformers == 1.2.0
numpy==1.21.6
pandas==1.3.5
Flask == 2.2.2 

2. 算法模型

  • 实体识别模型:BERT(wwm)-BiLSTM-CRF

  • 实体消歧模型:Jaro-Winkler Distance

  • 情绪分析模型:BERT(FT)-Pool-MLP

3. 运行方式

3.1 安装Python环境

3.2 下载辅助文件(包含bert-base-chineseoutput两个文件夹):https://drive.google.com/file/d/13IYKsJBv-2gNHqJZDfmPp5foeNDrGeJq/view?usp=sharing 解压后将这两个文件夹放到与项目同级的目录,具体结构如下:

- 项目根目录
  - Flask_Web
    - api.py
    - models.py
    - predict.py
    - ...
  - bert-base-chinese
    - config.json
    - ...
  - output
  	- pytorch_model.bin
  	- MLP_best_MSE.pth
  	- BERT_best_MSE.pth
  	- ...

3.3 进入Flask_Web目录启动项目

python api.py

项目展示

海报

项目海报

实验结果

实验结果

项目介绍页面

项目介绍

金融实体识别及情绪分析页面

金融实体识别与情绪分析1

金融实体识别与情绪分析2

关于我们页面

关于我们

About

Entry for the 19th "Citi Cup" Financial Innovation Application Competition 2024. Our submission, Fink_NER_ED_Web, is an extension of the Fink project for the 19th "Citi Cup" Financial Innovation Application Competition in 2024. Centered around the BERT model, it integrates two downstream tasks: entity extraction and disambiguation, as well as senti

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • SCSS 67.9%
  • CSS 11.6%
  • Python 9.6%
  • HTML 5.7%
  • JavaScript 5.2%