Visualizing conversational data in virtual reality

Abstract: Since the first implementation of a simple chatbot was made in1964, countless research and development have been made to makethe fascinating idea of talking to a computer a reality. But not untilrecently, have chatbots started to make an appearance in everydaylives amongst a broader audience. As the popularity of chatbotsincreases, the demands and functionality of the chatbots rises whichconsequently expands the size and complexity of the chatbot. Theconversational data from a chatbot can become very complex andhard to understand. Therefore, to ensure the continuousadvancement of features in chatbots, the developer needs tools andinstruments to compete in the growing market. Through a prototype based design process, a problem amongstdevelopers to visualize and understand the conversational data froma chatbot is first identified and addressed. A Conversational DataVisualization (CDV) prototype in virtual reality is then developedwith the intention to help developers understand and explore theconversational data from the chatbot they are working on. Thedesign of the CDV is based on theories about key features ofvisualizations in 3D and related work that study visualizations withsimilar data structures as the conversational data from chatbots.Furthermore, the features of the CDV is based on the identifiedproblem of visualizing conversational data amongst developers.Due to the importance of participatory design in a design process,an exploratory usability test of the CDV prototype was conductedto further explore the design choices regarding the identifiedproblem. The conversational data is visualized with tree structures in acircular formation to allow for visualization of links betweendifferent conversations. Results from the explorative usability testindicates that the visualization gave the users of the CDV anunderstandable overview of the conversational data. However,finding specific stories and nodes in the conversational data wasidentified as a problem due to inadequate information in theoverview of the visualization.