System for persona ensemble clustering: A cluster ensemble approach to persona development

The personas approach to user modeling aims to improve system interface design and increase the chances of information system success. Whereas there have been recent attempts to semi-automate the persona clustering process, the current methods fail to conduct simultaneous data analysis utilizing both quantitative and qualitative data; therefore, the current methods do not approximate human clustering judgment. Additionally, traditional manual persona clustering methods are resource intensive. This study views solutions to this problem from a design science lens in order to find utility in artifacts that meet the needs of an organization. Two information technology artifacts are built and evaluated: a new ensemble clustering method for persona development and an instantiation of the method in a prototype known as the System for Persona Ensemble Clustering. In order to validate the new method and prototype system, data are collected on system users in the context of a military Knowledge Management System. All data were simultaneously analyzed and combined using an ensemble cluster method implemented in the prototype system. The clustering effectiveness of three existing persona clustering methods was compared with the new method by using an expert panel clustering as the baseline. Final agreement measures for the three existing methods and the new method indicate the new semi-automated ensemble cluster method creates persona clusters more effectively. Additionally, the new method is more than three times faster than the traditional manual clustering method. These results suggest that ensemble clustering methods are effective at triangulating the plethora of user data commonly available for persona development projects. As the amount of user data available to design teams grows, it is imperative to use qualitative and quantitative data simultaneously to understand user goals, needs, and behaviors. The results of an effective and efficient persona clustering method are realistic personas that portray typical system users and system interfaces that match those personas. Finally, the research is communicated to management-oriented and technically-oriented audiences using a design science research perspective.