CBT-Driven Chatbot with Seq-to-Seq Model for Indian Languages

Abstract

Chatbots, also known as conversational interfaces, provide a new way for people to interact with computer systems. Chatbot software is a feature or an application for communicating directly through natural language with online users, so as to allow their queries to be solved without having to wait in a queue for a call center agent to pick up. In a way, chatbots are considered as virtual assistants for the user, with whom they can interact smoothly. Focusing on Covid-19, for example, we describe a cognitive–behavioural therapy (CBT)-driven chatbot in development which provides frequently asked questions about COVID-19 and relevant information resources. We emphasize the mental health aspect of the disease because Covid-19 not only impacts physical health but mental health as well. In response to these needs, our chatbot offers a cognitive Patient Health Questionnaire. Based on the user’s sentiment analysis score, the chatbot will then guide the user to find helpful resources to address mental health issues such as depression. Nevertheless, we encountered many challenges while building a chatbot in Indian languages: no appropriate data and information in Hindi language related to COVID-19 and mental health were available. Therefore, maintaining and adding data in the database was a tedious task. Being morphologically rich, free word order languages, Indian dialects carry ambiguities and complexities and are therefore complex to handle. Most of the Indian languages fall under the low-resource language category, as the datasets currently available for most Indian languages are limited when compared to the English language. Because the World Health Organization has declared Covid-19 as a public health emergency of international concern (PHEIC) making information about Covid-19 available to speakers of many languages is health-care priority. Guided by this initiative to help make information available in a public health crisis, we designed the chatbot by collecting relevant data from the user and then displaying the results in web and mobile devices. This CBT-driven chatbot is built using the transformer model architecture, which is based on the attention mechanism. Previously, the authors worked on this model for the English language, and now to advance the chatbot in different languages the authors are developing the chatbot in the Hindi language as well. We have applied a seq-to-seq transformer-based deep learning model into proposed work which presents an end-to-end potential in the domain. We found that the development of a chatbot in Hindi was necessary to help people who know only Hindi language to converse with the chatbot easily.