Korean Restaurant Reservation Dialogue System

Implement korean restaurant reservation dialogue system based on hybrid code network(https://github.com/johndpope/hcn).
Add post processing and edit templates to adjust korean dataset which we created. 759 training dialogue data and 190 test dialogue data were used for Korean restaurant reservation dialogue system model.
Experimental results show that the proposed system has 95% accuracy of per-response and 63% accuracy of per-dialogue.

Download Word2vec Trained with Korean Data

1. cd data/
2. Downalod word2vec
3. tar -xvf korean_word2vec.tar.gz

Train

python3 train.py

Building the Dialogue Corpus

The Korean dataset is consist of 1000 dialogues.(data/korean_train)

Restaurant Reservation System Data Translation

Variation of Speech Patterns according to Purpose of Utterance in Korean

System Architecture

Interaction

Result

Hyper-parameter

Evaluation

• Per-response Accuracy : 95%
  
• Per-dialogue Accuracy : 71%