Data mining library reuse patterns using generalized association rules

Abstract

In this paper, we show how data mining can be used to discover library reuse patterns in existing applications. Specifically, we consider the problem of discovering library classes and member functions that are typically reused in combination by application classes. This paper improves upon our earlier research using "association rules" [8] by taking into account the inheritance hierarchy using "generalized association rules". This turns out to be a non-trivial but worthwhile endeavor.By browsing generalized association rules, a developer can discover patterns in library usage in a way that takes into account inheritance relationships. For example, such a rule might tell us that application classes that inherit from a particular library class often instantiate another class or one of its descendents. We illustrate the approach using our tool, CodeWeb, by demonstrating characteristic ways in which applications reuse classes in the KDE application framework.

Comparing Approaches to Mining Source Code for Call-Usage Patterns

“…Michail [10] presented an approach based on itemset and association-rule mining to uncover entities such as components, classes, and functions that occur frequently together in library usages. Similar to the work presented here, Li et al [7] addresses the question of extracting rules and violations of typical usages of function calls in a system.…”

“…Recently, researchers [7,8,10] have applied datamining techniques, specifically frequent-pattern mining algorithms, to the problem of uncovering/discovering call-usage patterns from the source code of large systems. The result is a set of rules that describe frequently occurring call-usage patterns within a system.…”

“…The result is a set of rules that describe frequently occurring call-usage patterns within a system. This has been found to be potentially useful for tasks such as identification of standard library/API usages and fault location [7,8,10]. The techniques used are quite efficient; however they sometimes result in a large number of false positives, i.e., reporting potential errors where none actually occur.…”

Rekonstruktion von Software-Architekturen

“…Die Mustererkennung wird zum Teil mit einer anschließenden Filterung anhand dynamischer Information über das Protokoll kombiniert, um Verhaltenseigenschaften bei der Suche miteinzubeziehen [82]. Auch Data-Mining-Techniken kommen zum Einsatz [85].…”

An approach to mining call-usage patternswith syntactic context

“…That is, they form latent programming rules that seldom exist outside the minds of developers. Such rules have been found to be potentially useful for tasks such as identification of standard library/API usages and fault location [4][5][6]. Violations of these rules can be very difficult to uncover, report to issue-tracking systems, and fix unless the rules are explicitly documented.…”