Kobart Chat Persona Extraction V2

🚀 KoBART Persona Extraction

This is a model that can extract a user's persona from their daily chat conversations.

🚀 Quick Start

This model can extract a user's persona from their daily chat conversations. It uses the KoBART model and is trained on the AIHub Korean Multi-session Dialogue dataset.

✨ Features

• Persona Extraction: Extract a user's persona from their daily chat conversations.
• Model Upgrade: Compared with the previous version, it is trained with informal language data converted from formal language data.

📦 Installation

No installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

## Input ##
['안녕, 나는 20대 여자야. 

나는 봉사활동을 하는게 즐거워.', 

'응 국내 해외 가릴거 없이 봉사활동을 다녀.', 

'아. 나는 콜센터 상담원이야.', 

'응 해외 여행을 자주 가. 간 김에 봉사활동도 하고 그래.',

 '응. 여기 일은 별로 안해.',

 '여기는 야근이 따로 없더라고.',

 ' 나도 이만 일 해야겠네.']

## Output ##
나는 20대 여자이다 나는 봉사활동을 하는 게 즐겁다,

나는 콜센터 상담원이다,

나는 해외여행을 자주 간다,

나는 야근이 없다.

📚 Documentation

• Model Used: KoBART
• Dataset: AIHub Korean Multi-session Dialogue
• GitHub Repository: https://github.com/DonghaeSuh/persona_extraction_model
• Detailed Training Method: Blog

🔧 Technical Details

Difference from v-1

• NLPBada/kobart-chat-persona-extraction-v1
• The previous model was trained with formal language data, while this model is trained with informal language data converted from formal language data.
  • The following model was used for conversion: squarelike/korean-style-converter-6b

Validation Score

• BLEU: 0.4405 | Property | Recall | Precision | F1 | |----------|--------|-----------|----| | Rouge-1 | 0.5319 | 0.5817 | 0.5501 | | Rouge-2 | 0.3170 | 0.3388 | 0.3266 | | Rouge-l | 0.5282 | 0.5778 | 0.5463 |

📄 License

No license information is provided in the original document.