Exploratory Analysis of Usage Statistics of Dialogue Systems by Visualization

Abstract

It is important to analyze the usage statistics of dialogue systems in order to identify improvements that can increase user satisfaction. However, further beneficial information may be hidden among the information (such as logs and registered usage statistics in the database) that is not output as a result of the general analysis methods and can be discovered through exploratory analysis. To this end, this study proposes a method for the exploratory analysis of the usage statistics of dialogue systems using visualization to detect beneficial information, either intentionally or unintentionally. Specifically, usage statistics are visualized as network graphs showing the queries and transitions among them, and functions to extract particular visualized objects and identify characteristic queries are implemented to support the analysis. The results of this evaluation show that the proposed method can detect beneficial information through an exploratory analysis of visualized usage statistics, thereby demonstrating its effectiveness in the analysis of dialogue systems.