The evaluation of existing user interfaces is a challenging task, especially when it comes to the use of smartphone applications. It is particularly hard to identify frequent tasks performed by the users, given that many tasks are of a non-sequential nature and can be manifested and ordered in different event sequences. Designed as an add-on project to FUSAMI, the “Data Mining and Visualization of Android usage data” (DAVE) project explored the application of the Latent Dirichlet Allocation (LDA) algorithm to the domain of usage data. It was possible to demonstrate that the algorithm, although being originated in the domain of Natural Language Processing, is capable of capturing the latent non-sequential task models hidden in the click streams of a prototypical Android application. In the scope of the project, the visualization of non-sequential tasks was also investigated through the use of heat maps. The LDA algorithm is a suitable algorithm to detect latent non-sequential tasks and the discovered task models might give additional insight to HCI specialists targeting the improvement of an existing smartphone application.
Author: Albano Brito
Type: MSc thesis
Partner: Faculdade de Ciências da Universidade do Porto