Filipino-designed Q&A bot TalkingVac answers COVID-19 queries

A team of local junior data scientists have developed a question-answering bot trained to respond to questions about COVID-19 vaccines.

Since the Philippine government launched its COVID-19 vaccination program in March, the team figured that Filipinos would have endless concerns that could result in vaccine hesitancy if left unanswered.

Only 770,751 COVID-19 vaccine doses have been administered in the Philippines as of April 1, according to the World Health Organization.

Aside from logistical issues, the slow progress was attributed to vaccine hesitancy among the Filipino population. A Pulse Asia survey conducted from February 22 to March 3 showed that 61% of respondents would say “no” to inoculation with a COVID-19 vaccine if available during the polling period. 

This is why the team of Phoemela Ballaran, Fatima Santos, Mikee Sevilla, and Kaye Yao created “TalkingVac,” a bot that allows users to ask a COVID-19 or vaccine-related question and receive an instant response.

TalkingVac currently knows over 300 question-and-answer pairs, with information lifted from official sources like the Department of Health and the World Health Organization. When a user asks a question, the bot matches this with a question it has in its database, then returns the answer.

This is done through a series of machine learning libraries that enable the bot to find the closest match for the user’s input.

To guide the building of the bot, the team first conducted sentiment analysis and topic-modeling on tweets relating to COVID-19 vaccines. Among the frequently asked questions were the efficacy, safety, and side effects of vaccines, as well as general guidelines on how to get vaccinated and what to do before and after vaccination.

“With the rollout of COVID vaccines here in the Philippines, TalkingVac aims to be able to provide a simple way to access information on COVID vaccines instead of going through different websites to search for answers. This is especially important given that it can be difficult for individuals to properly parse through websites and identify reliable information,” the team said.

In development

Presently, the bot is still in development and has some limitations. It can’t provide real-time COVID-19 statistics nor can it answer very specific, personalized, or localized vaccine information. Additionally, it can only respond to questions in English.

The team hopes to expand the capabilities of TalkingVac, but they said the effort requires assistance or collaboration with the government and health sector in order to properly execute the desired improvements. Providing localized information, for example, entails close coordination with respective local health units to determine where the bot can best help

“We noticed as well that there are quite a handful of local government units and organizations using online platforms to manage vaccination. We feel that these kinds of online platforms for vaccination might be a good place to integrate TalkingVac. Once a user has accessed this platform, it means that the user is already interested in vaccination and would likely be seeking more information about it online,” the team said.

TalkingVac’s current design is best utilized for answering general concerns like vaccine guidelines, dosage, and brands all in one application. It currently has two versions: the first uses ChatterBot (a Python-based conversational dialog engine) and the second leverages the BERT model (developed and used by Google in their search engine). 

Chatterbot is best with conversation-style exchanges while BERT is a state-of-the-art language model. Anyone can access the said Q&A bots here. Simply click on the “QABot” option on the sidebar and you will be directed to a page where you can access the relevant links.

TalkingVac was launched during the team’s April 13 demo day showcase for online data upskilling school Eskwelabs. The team had met through the Eskwelabs Data Science Fellowship, a 12-week program for aspiring data scientists where the final requirement is a capstone project that encompasses everything learned during the bootcamp.

“For our capstone, our group decided that we wanted to pursue a ‘Data For Good’ project so it could benefit Filipinos,” the team, which was given 3 weeks to design and execute their project, said. They found it challenging to scrape and clean Twitter data, as well as to compile the question-and-answer dataset on COVID-19 vaccines from credible web sources.

Although the team did not receive any funding for the TalkingVac project, they were guided by their Eskwelabs mentor Albert Yumol in applying all the data science techniques and applications learning under the fellowship. “The tools we used are open-source so aside from time and energy, we didn’t really spend anything for the project,” the team said. – Rappler.com