Semi-automatic Aggregation of Multiple Models of Visual Attention for Model-Based User Interface Evaluation

Predicting the distribution of attention to new user interface designs can provide valuable information during the design process, but accurate predictions are difficult to achieve. Recent studies have shown that accuracy can be increased based on the Diversity Prediction Theorem if multiple, independently developed models for the prediction of attention distribution are aggregated. However, aggregating multiple models is a manual task, that takes a lot of effort because a large number of information sources, which are defined as parts of each model, need to be compared among each other. In this work we test two different clustering approaches for automatically aggregating such models. We show that the clustering quality is not sufficient for fully automatic clustering and present a software-supported solution for a semi-automatic clustering process.