Representativeness and Descriptiveness of Task Trees Generated from Website Usage Traces

Abstract

Task trees are often used to define the actions on a software as well as their order which is required to accomplish a certain task. With an increasing task complexity, their creation can be laborious and error-prone. Hence, there was work done to generate them automatically from recordings of user actions. In this paper, we assess for one of these approaches if the generated task trees are representative and descriptive for recorded and also unrecorded user actions. This characteristic is important as it allows for subsequent valid analyses of the software usage based on these task trees. For our evaluations, we transform the task trees generated from one set of recorded actions into grammars for the language spoken between the user and the software. From these grammars, we generate parsers with which we try to parse action combinations in other usage recordings. Our results show, that the approach under analysis produces partially representative task trees, which are also descriptive for unrecorded user behavior.