Enhancing Feature Location Accuracy Using Textual and Structural Extraction

Some programmers have difficulty when they are debugging or developing a new feature. Program comprehension is an activity to understand the behavior of a specific code. A prior understanding of a code could help the performance of a programmer. The feature location is the activity to identify which part of the source code correlated with a specific feature. The feature location mostly uses information retrieval and text processing. The usage of known structure was limited to the model on a specific domain. The proposed method combines textual-based information retrieval and class structure extraction (TESA) of elements of codes to help increase the precision and recall of feature location. The textual processing includes natural language processing and indexing using Vector Space Model (VSM)-Lucene to determine what the best token as the query. The structural extraction process was done by extracting the class member from the class relationship to determine where the correct class to expand the direction. The class relationship is used as the base of searching feature location expansion. The dataset was based on Java and using Model View Controller (MVC) design. The proposed method achieved 91% of precision and 95% of recall.