Mining API Usage Patterns Based On Closed Partial Order

Application Programming Interface(API)is the most commonly used modular approach in nowadays software development.As the software development market grows larger,more and more APIs are being developed and used.However,most API use specifications are not specified in the API reference documentation,in that case,programmers need to search relevant code from the Internet to select and use.With a large amount of source code available,it has become a hot issue to mining the API usage pattern from the source code to help programmers achieve fast and efficient development.In this thesis,poset is used for module sentence sequences,and the algorithm and its mining effect on API corpus are studied from the perspective of frequent closed partial order pattern mining.The main tasks include:(1)A benchmark corpus for API was established.A benchmark corpus for comparing API mining algorithms was established due to the selected source files with API usage examples from the open source project corpus.The problem,that mining results were not compared with standard data sets,was solved because the code sample(example)was used as a positive example.(2)The benchmark corpus were preprocessed to represent the source code as sequences.A sequence similarity calculation method based on N-gram was proposed and hierarchical clustering was carried out for API method call sequences,which solved the influence of different usage scenarios on sequence support.(3)A Group Based Closed Partial Order(GCPO)mining algorithm was proposed to model the source code statement sequence with partial ordered set.The experimental results were evaluated by several indexes including accuracy rate,conciseness,coverage and algorithm running time,and were compared with CLAMS algorithm,UPMiner algorithm and other three algorithms,which fully verified the effectiveness of the API usage pattern mining method proposed in this thesis.