실시간 대화 요약 & 컨텐츠 추천 봇: TravelTalk

1. Introduction

Members

김현수	이성구	이현준	조문기	조익노

Contribution

• 김현수   Front/Back-end 구현 • ElasticSearch 구현
• 이성구   Product Manager(PM) • Dialogue summarization 고도화
• 이현준   Dense retriever baseline 학습 및 평가 • ICT 응용기법 학습 및 평가 데이터 제작
• 조문기   Front/Back-end 구현 • Dialogue summarization 및 metric 구현
• 조익노   Data crawling scheduler 구현 • MongoDB 적용 • Retriever model 구현

Project Tree

|-- app
|   |-- assets
|   |-- src
|   |   |-- elastic
|   |   └-- models
|   |-- templates
|   |-- app.py
|   |-- config.py
|   └-- mongodb.py
|-- train
|   |-- summary
|   └-- retriever
└-- monstache
    └-- mongo-elastic.toml

2. Model

KoBART

Result

ColBERT

Result

3. Flow Chart

System Architecture

Pipeline

4. How to Use

Install Requirements

pip install -r requirements.txt

Getting Started

• Summary model train

python ./train/summary/train.py

• Retriever model train

python ./train/retriever/train.py

• Run chat app with model server

python ./app.py
python ./src/models/summary_model.py
python ./src/models/retriever_model.py

5. Demo

• Video Link: YouTube

6. Reference

• Dense Passage Retrieval for Open-Domain Question Answering
• Latent Retrieval for Weakly Supervised Open Domain Question Answering
• ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT
• 네이버 ColBERT 사용방법
• Reference and Document Aware Semantic Evaluation Methods for Korean Language Summarization
• Better Fine-Tuning by Reducing Representational Collapse
• Momentum Calibration for Text Generation
• 요약 관련 Survey 자료 : https://github.com/uoneway/Text-Summarization-Repo
• 요약 모델 (KoBART) baseline : https://github.com/seujung/KoBART-summarization
• 요약 모델 성능 개선 방법 (Scatterlab) : https://tech.scatterlab.co.kr/alaggung-dlaggung-dialog-summary/