Mining balanced API protocols

摘要

API protocols can be used in many aspects of software engineering, such as software testing, program validation, software documentation, etc. Mining API protocols based on probabilistic models is proved to be an effective approach to achieve protocols automatically. However, it always achieves unbalanced protocols, that is, protocols described using probabilistic models have unexpected extremely high and low probabilities. In this paper, we discuss the unbalanced probability problem and propose to address it by preprocessing method call sequences used for training. Our method first finds tandem arrays in method call sequences based on the suffix tree. Then, it substitutes each tandem array with a tandem repeat. Since repeated sub method call sequences are eliminated, balanced API protocols may be achieved. In order to investigate the feasibility and effectiveness of our approach, we implemented it in our previous prototype tool ISpecMiner and used the tool to perform a comparison test based on several real-world applications. Experimental results show that our approach can achieve more balanced API protocols than existing approaches, which is essential for mining valid and precise API protocols.