An open QA system, for Romanian, answering open-ended questions about COVID-19. Works by interrogating Bing with a Romanian question to query algorithm, returning the top 10 text snippets. Using a COVID-19 fine-tuned BERT model, the most probable answer(s) are sent back to the user.
You need a Bing V7 search API key that is valid and that is stored in the environment variable BING_SEARCH_V7_KEY.
The main entrypoint in the Python 3 code is covidqa.covid_qa_system() method which takes a str question and returns a list of dictionaries, each with probable answer information. A dictionary from this list looks like this:
{
"name": "Infectarea cu COVID-19 în timpul sarcinii îți poate afecta ...",
"url": "https://doc.ro/sanatate/infectarea-cu-covid-19-in-timpul-sarcinii-iti-poate-afecta-bebelusul",
"snippet": "Tratamentul pentru COVID-19 este similar cu tratamentul altor boli respiratorii. Indiferent dacă ești însărcinată sau nu, medicii recomandă: acetaminofen (Paracetamol) pentru o febră de 38°C sau mai mare hidratare cu apă sau băuturi cu conținut scăzut de zahăr odihnă",
"date": "2022-03-19T19:53:00.0000000Z",
"answer": "Tratamentul pentru COVID-19 este similar cu tratamentul altor boli respiratorii.",
"confidence": 0.9542,
"start_offset": 0,
"end_offset": 80
}Developed in the Enrich4All project, for Romanian COVID-19 questions. To be extended to new languages and domains.