Introduction

In Early 2020, as the World was just beginning to understand the occurrence of a pandemic, growing misinformation and lack of immediate self-assessment was a major problem, in India. We, a team of 3 CS students, identified the need of a one - stop interactive, yet effective solution for all things COVID-19, that uses real-time data collection, trusted Covid-19 databases and Machine Learning algorithms for reporting and Self Assessment tests : A Covid-19 Assistance Chatbot.

Executive Summary

What?

It’s a COVID 19 Assistance Chatbot that combines NLP and ML Algorithms and has following functionality :

• COVID 19 Self Assessment Test
• Answers FAQs from trusted sources
• Reports Real-time, region specific live case statistics & Vaccination updates.
• Nearest Hospital locator and real-time availability

The Design Process

Empathise

Who are our users? What do they need? What are they struggling with?

User Interviews

Ten users of varied age groups were interviewed to understand the pain-points and challenges they encounter while looking for COVID-19 related information through various sources.

Define

Then the accumulated insights & pain-points from the user research were turned into actionable objectives in order to define the scope of the project.

User Insights to Actionable Project Goals :

Data Collection

Validated, crowd-sourced datasets were found for COVID-19 Likelihood, live case statistics.

Sample Dataset I : Dataset with 20000 entries, that detailed exhaustive list of age, symptoms that relate with having COVID-19.

Ideation & Synthesis

Detailed mind-maps were created to lay down various options for available resources to make the chatbot, as a result of group brain-storming sessions.

Furthermore, a comparative accuracy analysis was done to determine ML Model most suitable for our use-case of self-assessment. I was solely responsible for performing this analysis.

Based on the available number of training sets(data)/features, we decided to proceed with Logistic Regression Model.

Development

Gathered insights and research were synthesized to define the conversational flow of the chatbot:

Defining the constituents of a “Good Conversation”

Human language is inherently cooperative and can operate on imaginary fields. The desirable characteristics of a good conversation were defined as follows:

• Polite
• Co-operative
• Goal-Oriented
• Context-aware
• Error-tolerant
• Turn-based

Designing AI Assistant Personality : CoviCare

Keeping in mind the subject of the Bot, the persona of the Bot must be calm and soothing, but not super-serious or erratic.

Based on the above, the utterances, Intents and Entities were defined for the Bot. A polite and concise dialogue was constructed.

Sample Dialogue Flow Digraph for a happy path scenario where user wants to say find hospitals near the, and also try the self assessment feature:

Code Snippets :

I was solely responsible for implementing the self-assessment and live-statistics features, while my group mates took up the other features.

Intents with sample utterances:

Deliver

Some Snapshots of the the CoviCare Chatbot

Rigorous testing was done to avoid echoing, enhance context awareness, and to make it error-tolerant.

Impact & Feedback

• User base of over 45 people, received positive feedback which was also used to make modifications to the Bot, so as to keep it evolving continuously.
• Users reported having found the Self-Assessment helpful, as a preliminary step before getting a Covid test.
• Was able to handle commonly asked questions, thus successfully countering spreading of misinformation.
• Live statistics and information was updated every few hours.
• Displaying nearby hospitals, based on availability had also proven to be effective.