Extensions to Mining Framework Annotation Rules

Framework usage is challenging because the requirements for the correctness are often implicit. We focus on making

such requirements more explicit by association rule mining on the data from client projects that use a framework.

We present an extension to an existing baseline method that does this. In particular, we examine alternative rule

quality measures used in the ranking of association rules mined, and alternatives in the selection of client projects.

Such alternatives are novel and have not been explored in the context of the baseline method. We evaluate the alternatives

by comparing their results to those produced by the baseline method. More concretely, we base the comparison on their

ranking of incorrect rules, and on their measurements for the Area Under Curve metric. We conclude that some of

the evaluated quality measures outperform the baseline for the ranking and selection of rules. We also show that the

selection of secondary client projects, adding some clients that do not directly use the framework of interest, matters.