roytravel / korean-chatbot

Task-oriented Korean dialogue system

Korean Chatbot

Korean chatbot is a task-oriented dialogue system. it orients a multi-domain, multi-turn and a hybrid(ontology-free and pre-defined ontology) dialogue system.

Environment and Installation

• Windows 10
• CUDA 11.2
• python 3.6
• pip install -r requirements.txt

Model Training

Run following three scripts to fine-tune the model.

>>> python train_intent.py
>>> python train_entity.py
>>> python train_question.py
>>> python train_summary.py

if you run a script, then it makes fine-tuned model. BERT model is used to train an intent classification, entity recognition model and question answering model for now. other model will be updated.

Model Inference

if you want to do inference, then you can run the following script. import Predict class from predict.py then create an instance method with a sentence that you want to query.

>>> from predict import Predict
    
>>> P = Predict()
    
>>> intent = P.predict_intent(sentence)
>>> entity = P.predict_entity(sentence)
>>> answer = P.predict_question(context, query)
>>> sents = P.predict_summary(sentences)

Server and Client

Run following script to run both server and client.

>>> python server.py
>>> python client.py

if you want to do a test, then you can type http://127.0.0.1:4999/?query=오늘 날씨 알려줘. after typing this, you can hear a voice and get information that is made by a bot. so far only the weather function is working.