We ran our tool on the dataset described in Figure 9. Dataset. Mined preconditions are compared with the ground-truth to determine the accuracy of the result in terms of precision and recall. We demonstrate the result for overall approach in this section. The usage-based mining approach without using implicit beliefs (Nguyen et al. 2014a) is used as the base case to show the absolute and relative improvement we achieve. The components for inferring implicit beliefs added to traditional usage based mining are the following: object instance creation, type comparison, null dereference, count-controlled loop, short circuit evaluation and local exception.
Figure 10 shows the absolute values and relative improvements in precision and recall for 7 libraries. Our approach achieved precision from 21%–88% and recall from 39%–100%. Overall, the precision and recall of our approach are high: 60% and 69%, respectively. The accuracy was improved on all libraries and overall by 32% in precision and 78% in recall. The result is also shown in following table format: