ABSTRACT
Cross-cultural interactions within the design team (and/or with external participants, as in co-design processes) can impact the team in many ways. In modeling changes in design processes over time, it is important to capture changes in key aspects of design processes, such as understanding of the problem and emergence of crucial new ideas. However, modeling such foundational change can be very effort-intensive for researchers. In this short paper, we explore the use of computational linguistic methods – such as topic modeling – as a cost-effective supplement to traditional methods in providing a “quick and dirty” birds-eye view of what designers are talking about, and how those topics change over time. Specifically, we use Latent Dirichlet Allocation – a foundational topic model approach – to explore changes in the diversity and prominence of topics over time in the DTRS11 dataset. We test the impacts of variations in key modeling assumptions and parameters on the quality of the resulting model. Overall, our topic models identify a robust shift in diversity of topics following the second co-creation session. We also find significant changes (both decreases and increases) in the prominence of several topics. These high-level patterns provide a quantitative complement to qualitative intuitions, and raise interesting new research questions (e.g., did the design team learn more from the second co-creation session compared to the first? If so, what changed?). In summary, our analysis demonstrates the benefits and potential limitations of computational linguistic methods as a supplement to traditional in-depth qualitative analysis of design processes.
