Taskdict

For the empirical analysis, the authors created two datasets: dataset1 includes all change tasks<br> that have at least one change set associated with them; dataset 2 is a subset of dataset 1 and<br> includes all change tasks that have at least one change set and one task context associated. For<br> the AspectJ project the authors only created dataset 1, since only two tasks had a task context<br> associated. For each project the authors split each dataset into a training set to create the<br> dictionary and a test set. For reasonable training and test sets and to be able to compare them<br> across projects, they chose the test set size to be ten. Also, to be able to compare the results<br> of the dictionaries using dierent in-formation sources (change sets and/or task contexts) as well<br> as preserving the chronological order, they chose the last ten change tasks in dataset 2 as the<br> test set for Mylyn.Tasks,Mylyn. Context and Remus and the last ten change tasks in dataset 1 for<br> AspectJ. All change tasks that were resolved before the ones in the test set were used as the<br> training set.