On January 31, 2020, WHO officially announced the coronavirus outbreak in China and other countries as a Public Health Emergency of International Concern. Immediately, Vietnam Ministry of Health had employed various methods to counter the disease, which may incur an outbreak at any time. Among those, one of the most important tasks is to bring quick and accurate situational updates to citizens to prevent public panic as a consequence to fake news. FPT Corporation had had the honor to accompany the Electronic Health Administration – Ministry of Health in building the Covid-19 prevention virtual assistant, and had finished 90% of its work in only 2 days.

So, how was the Covid-19 prevention virtual assistant so quick?

1. Identify the problem and purpose of bot

After receiving the problem from the Electronic Health Administration – Ministry of Health on February 1, the FPT.AI team had immediately started to discuss and plan for building a chatbot – the Covid-19 prevention virtual assistant. The chatbot was to be responsible for answering inquiries and provide new information regarding the Covid-19 outbreak status. Aside from familiar mainstream medias, the virtual assistant will also be a mainstream news provider from the Ministry of Health, with quick information that is constantly updated in real-time. Information provided by the chatbot are gathered from specialists and doctors at the National Hospital of Tropical Diseases. Therefore, people can fully rely on it for accurate information on prevention and diagnosis instructions.

2. Prepare materials, intentions, answers, and samples

In order to have the most accurate information for bot-building, the FPT.AI team had to gather materials from various relevant parties. In particular, the Ministry of Health provides official correspondences, recommendations, guidance, and statistics on diagnosed cases, recoveries, and deaths domestically as well as internally. While information regarding the coronavirus, symptoms, prevention measures… are provided by doctors and specialists at the National Hospital of Tropical Diseases in consideration and great details. The Ministry of Health also provide in details lists of local hospitals ready for dealing with suspected cases, and requirements in self-quarantine.

3. Identify main topics of interest

After gathering enough information regarding the Covid-19 outbreak that are helpful to people, the Covid-19 prevention virtual assistant team had to identify problems that users are interested in. From missions of the bot, the bot builders had to find main topics on Covid-19.

For example, to build a chatbot on covid-19, we need to know what are topics of interest for users? They may care about information like what type of virus is the covid-19, where did it originate from, how does it spread, what are the characteristics of it…; and information like how to know if you are infected with the virus, what procedures must be taken in case of suspected infection, what are the fees of testing and treatment…; or information on the disease’s symptoms, prevention measures, status of the covid-19 outbreak around the world as well as in Vietnam…

For comprehensive information coverage in the chatbot, the bot team had to list as many topics of interest as possible. Then, the most effective approach was to make diagrams of main conversation topics, then arrange information in tree charts, from generals to details. This would help the chatbot get information in large quantities, in details, meticulously, and in full.

Under strong spirits to fight the disease, the Covid-19 prevention virtual assistant team had finished one of the most basic and fundamental steps in bot building in only 1 day – identifying users’ topics of interest. In part 2, we will share with your detailed steps to deploy the bot on the FPT.AI platform, including:

  • Real-time update of the outbreak’s status using the Json API card
  • Building a Training database from preexisting FAQs
  • Improve the chatbot quality using past conversations with users.

To be continue

FPT.AI

Related posts: