SMArTIC: Specification Mining Architecture with Trace fIltering and Clustering

Improper management of software evolution commonly leads to a lack of up-to-date specification. This situation is further aggravated by imprecise, changing requirements and short time to market requirement, which can result in software that is characterized by presence of bugs, anomalies and even security threat. Software specification mining is a new technique to address this concern by inferring specifications automatically. In this paper, we propose a novel specification mining architecture called SMArTIC (Specification Mining Architecture with Trace fIltering and Clustering) to improve the accuracy, robustness and scalability of specification miners. This architecture is constructed based on two hypotheses: (1) Erroneous transactions should be pruned from traces input to a miner, and (2) Clustering related traces will localize inaccuracies in learning and reduce overgeneralization. Corresponding, SMArTIC comprises four components: an erroneous-trace filtering block, a relatedtrace clustering block, a learner, and a merger. We show through experiment that the quality of specification miner can be significantly improved using SMArTIC.