An open source chatbot for a complex domain

Abstract

The call centre industry has grown rapidly due to advancements in information and communication technology that have enabled automation of many customer service tasks. The computerisation of customer relations has become a primary objective for many businesses, with the goals of generating productivity gains and establishing strong customer relationships. However, the high-pressure work environment of call centres often leads to high levels of stress for agents and high staff turnover rates. AI technology is bringing significant changes to call centres with intelligent call routing and analysis of customer behaviour allowing staff to handle calls more efficiently thereby reducing customer wait times. Chatbots are being deployed in call centres in order to improve the customer experience by handling routine queries efficiently and reducing the service agent workload. This research explores the chatbot built using open source technology to address customer queries for a complex domain, namely health insurance. This domain has technical terminology and products with detailed specifications. A knowledge base is designed and built to extend the existing domain taxonomy to cater for language and concepts used by ordinary users. A dedicated quiz application is developed to get real time feedback from users as they interact with the chatbot. The experimental results and evaluation presented in this work corroborate the idea that a chatbot can be applied to a complex domain. An initial survey indicates the complexity of the domain both in terms of the vocabulary and also the products. A chatbot built using the Rasa open source framework is able to present complex data from the domain in a manner that users can understand. An iterative process is used to modify the chatbot using open questions posed to participants and analysis of the language used by participants, resulting in improved understanding of user requests and better usability.