ODQA-Demo-Site v2.1

Architecture

• Guideline : Explain how to use it for 20 seconds for the first user.
• Preprocess question : After the question is entered, correct the spelling.
• Daily chat : Rule-based daily conversation
• Short question : For short questions of less than two syllables, it answers wiki url by extracting keywords from those questions.
• Expected Q&A : If there is a highly similar expected question, output the answer to the expected question.
• Retrieval : If it is impossible to retrieve a document, reply that "I don't understand your question.".
• Reader : If the MRC's answer does not exceed the standard score, chatbot says "I'm not sure.".
• Postprocess answer : In the case of Korean, there is an investigation, so it is removed.

Usage

$> tree -d
.
├── static
│     ├── css
│     ├── image
│     └── js
├── templates
├── app.py
├── dpr_inference.py
├── mrc_inference.py
├── tag_inference.py
├── Retrieval.ipynb
└── Setting.ipynb

1. Setting.ipynb : Install the required libraries.
2. Retrieval.ipynb : Set Elasticsearch.
3. app.py : Load the required modules, then run flask.