Brief Review — A Proposed Chatbot Framework for COVID-19

COVID-19 Chatbot Using BERT

Sik-Ho Tsang
3 min readApr 2, 2024
BERT

A Proposed Chatbot Framework for COVID-19
COVID-19 Chatbot Using BERT
, by Misr International University
2021 MIUCC, Over 30 Citations (Sik-Ho Tsang @ Medium)

Medical/Clinical/Healthcare NLP/LLM
20172023 [MultiMedQA, HealthSearchQA, Med-PaLM] [Med-PaLM 2] [GPT-4 in Radiology] [ChatGPT & GPT‑4 on USMLE] [Regulatory Oversight of LLM] [ExBEHRT] [ChatDoctor] [DoctorGLM]
==== My Other Paper Readings Are Also Over Here ====

  • A smart chatbot system that can communicate with people and provide them with answers about the COVID-19.
  • The first step is a text classification technique that employs the BERT Transformer to categorise text input into various categories based on the meaning of the words themselves.
  • The actual application of the BERT model, as well as the query domain for answers, is the second step.

Outline

  1. COVID-19 Chatbot Using BERT
  2. Results

1. COVID-19 Chatbot Using BERT

1.1. Pre-Processing

  • The major steps for data pre-processing are summarized below:
  1. Removal of unnecessary spaces and null values through regular expressions.
  2. Converting all text into lowercase.
  3. Customizing the length of questions and answers.

1.2. Text Classification Using BERT

  • The number of text strings that is allowed by BERT is limited (usually 512 tokens). The responses are divided into categories based on the type of question. As a result, based on the extracted keywords or the meaning of the given query, the BERT classifier is used to decide which response category to use.
  • The augmented dataset [30] consists of 4115 questions all of them are categorized into 14 different categories which are ‘Speculation’, ’Transmission’, ‘Nomenclature’, ‘Reporting’, ‘Societal Response’, ‘Societal Effects’, ‘Origin’, ‘Prevention’, ‘Treatment’, ‘Testing’, ‘Comparison’, ‘Economic Effects’, ‘Symptoms’, ‘Having COVID’, ‘Individual Response’.
  • 75% training, 25% testing.

The BERT model classification showed training accuracy of 98% and testing accuracy of 96%.

1.3. Answering Questions Using BERT

Chatbot
  • The BERT model is fed questions as input and then returns answers based on the context given.

2. Results

Sample Results
  • The above table provides some questions along with the system answer.

The proposed model showed reliable answers for common questions that users frequently ask regarding COVID-19.

  • There is no accuracy provided for question answering (QA), but authors mentioned that it is still a preliminary version, they are still developing to ensure the robustness.

--

--

Sik-Ho Tsang

PhD, Researcher. I share what I learn. :) Linktree: https://linktr.ee/shtsang for Twitter, LinkedIn, etc.