Multi-layered Approach for Recovering Links between
Bug Reports and Fixes

Anh Tuan Nguyen, Tung Thanh Nguyen, Hoan Anh Nguyen, Tien N. Nguyen


Empirical Evaluation


 Data Collection and Evaluation Metrics

For comparison, we used the same data set as in ReLink in which the ground truth for bug-to-fix links of three projects are publicly provided at .

Subject Systems

Table 1. Subject Systems

Table 1 summaries their information with time and revision ranges, the number of fixed bugs, and the number of links. We also downloaded the respective bug records (summaries, descriptions, com-ments, and meta-data) as well as the commit logs, changed source code, and meta-data from the projects' Web sites. The recovered links from MLink were compared to the ground truth. As in Relink, we also use three evaluation metrics.Precisionis defined as the ratio between the total number of correctly detected links and the total number of detected links. Recallis the ratio between the number of correctly detected links and the total links in the ground truth. F-scoreis balanced metric for precision and recall and defined as

F-score =2*Precision*Recall/(Precision+Recall)

Accuracy of Combinations of Modules

Accuracy Result for Cascading Layers

Table 2. Accuracy Result for Cascading Layers


Comparisons with ReLink and Topic Model

Comparison of Measurement Data

Table 3. Comparison of Measurement Data