This project represents the idea of typical QA system nowadays. It combines the idea of retrieval models and latest QA solutions.
It builds website-specific index to search on top of and provides precise answers for the end-user.
- Allow user to search for information on a website via native language.
- The user expects precise answer on his question.
- The system must suggest a link to the page the information was found on.
- Install
requirements.txt
- Download
en-core-web-md
SpaCy dataset viapython -m spacy download en_core_web_md
- Install package locally
pip install -e .
- Start server
FLASK_APP=server flask run
- Install google chrome extension from local files
- Open google chrome.
- Go to
Settings -> Extensions
- Enable
Developer mode
- Click
Load unpacked
and select folderclient
- Open new tab and click on the extension, write queries about cooking.
- Queries w/o
?
mark in the end will be treated as a search for matching recipe (Ctrl-click one of them). - Query with
?
mark in the end are Question-Answer request and the answer will appear in the same block.
- Queries w/o
- Apply steps 1-3 from previous instructions
- ...
- Add Question-Answering model for Russian language (DeepPavlov does not have one).
- Fine-tune model for specific area of interest (e-commerce, company-products, etc.)
- Create website-plugin instead of chrome extension to be natively loaded with the website.
- Replace tf-idf with other, more production-ready tools.