Lecture: Implementing a Chatbot - Rule Based vs. Machine Learning
Tools for implementing chatbots can generally be divided into two categories - rule based and machine learning systems. This presentation is about a bachelor thesis project that compared the implementation process of a customer service chatbot using two competing tools and a user study in which the resulting were chatbots were evaluated.
Customer service chatbots can be implemented using either a rule-based system or one based on machine learning. My bachelor thesis compares the system developed by Kauz Linguistic Technologies (an example of a rule-based system) with Dialogflow by Google (an example of a machine learning-based system) from both a developer perspective and the perspective of the end user. Each system was used to implement a German-language chatbot for the use case of a consultation for a bank account. The platforms were then evaluated using a systematic framework. In a pilot user study, 20 participants (mean age of 23, 45% female and 55% male) evaluated the resulting chatbots in terms of usability, usefulness and satisfaction. With regard to the implementation process, the Kauz system was found to have more built-in functionality as well as data structures that make it easy to make changes to the finished chatbot while Dialogflow was easier to troubleshoot and made it simpler to extract and remember information from user queries. In the user study, while ratings were generally positive for both chatbots, participants showed a preference for the chatbot developed with the Kauz system.