Simple Tracelink Discovery (STD) is a simple approach for Traceability Link Recovery. Its intended use is for linking informal textual software architecture documentation and software architecture models, i.e., PCM.
STD calculates trace links between a software documentation and a model by matching the model entity names and its n-grams with words and n-grams in the documentation lines.
For instance, for the entity AutomatedActionFactory, the matches of "automatedactionfactory", "automated action", "action factory" and "automated action factory" in each documentation line are counted. If the number of matches exceeds a certain threshold (match number threshold), a trace link between this entity and the corresponding documentation line is created.
For string comparison to match n-grams or entity names, the Levenshtein Distance is used with a simple normalization to the range [0,1]. If this value is greater than or equal to a defined similarity threshold, a match is counted. A threshold of 1.0 means that both strings or n-grams must be identical. Cases are always ignored. Analogously, the Jaro-Winkler Similarity can be used instead of the normalized Levenshtein Distance.
We evaluated the traceability link recovery on five open source projects. See also ArDoCo/Benchmark. The table below shows the results with a match count threshold of 1 and a similarity threshold of 0.9 for the normalized Levenshtein Distance.
| Project | precision | recall | F1-Score | Accuracy | Specificity | Phi | PhiNorm |
|---|---|---|---|---|---|---|---|
| Media Store | 1.000 | 0.383 | 0.554 | 0.932 | 1.000 | 0.596 | 1.000 |
| TeaStore | 0.882 | 0.357 | 0.508 | 0.908 | 0.993 | 0.526 | 0.864 |
| TEAMMATES | 0.759 | 0.446 | 0.562 | 0.861 | 0.965 | 0.510 | 0.699 |
| BigBlueButton | 0.375 | 0.148 | 0.212 | 0.897 | 0.974 | 0.189 | 0.310 |
| JabRef | 0.867 | 0.419 | 0.565 | 0.837 | 0.978 | 0.528 | 0.822 |