Extracting conceptual interoperability constraints from API documentation using machine learning

Abstract

Successfully using a software web-service/platform API requires satisfying its conceptual interoperability constraints that are stated within its shared documentation. However, manual and unguided analysis of text in API documents is a tedious and time consuming task. In this work, we present our empirical-based methodology of using machine learning techniques for automatically identifying conceptual interoperability constraints from natural language text. We also show some initial promising results of our research

Using software testing to repair models

“…In order to test our approach we conducted the following two experiments 8 . In the first experiment, we applied mutation to a set of models taken from the literature.…”

“…In model-driven engineering, there are approaches aiming at repairing inconsistencies between models [145]; Tran et al show a way to address the problem manually for the Linux Kernel [202]; Nadi et al present a method to automatically mine conditions under which a system behaves in a certain way [154], and there are methods to statistically infer constraints from data [61,8], but they are not directly applicable to repair existing models made of sets of constraints and they do not guarantee complete accuracy. A quality-based model refactoring framework as-Chapter 2.…”

“…Combinatorial models may have also constraints among parameter values to, for example, model inconsistencies between certain hardware components, limitations of the possible system configurations, or simply because of design choices. Some methods have been introduced to automate the process of inferring constraints, but they do not aim at repairing existing ones [8,197]. Therefore, in [88] we proposed an iterative approach that uses a fault-localization tool based on combinatorial testing, called BEN, and CIT policies introduced in our previous work [89] to find failure-inducing combinations of parameter values.…”

“…Some approaches that aim at simplifying complex constraints exist [120,206], but they may diminish readability and other qualities, such as compactness, traceability, and maintainability. 8 In this section, we propose a heuristic approach that tries to achieve the update request as much as possible, by using the target as oracle to compute the fault ratio. The fault ratio tells how close we are to the correct solution.…”

“…There exist methods to statistically infer constraints from sampled configurations [61,8,195,196,197,16]: they use a classifier to infer the conditions among parameter values, that determine a particular property, either a parameter above/below a certain threshold (like in [196]), or directly the configuration being accepted or rejected by the system (as in [197]). These machine-learning based methods are well-documented and supported by application studies to real scenarios, such as learning constraints among parameters in SCAD programs that may cause defects of configurable objects to 3D print [16]; and, in the case of L A T E X, showing that it is possible to obtain constraints among Boolean or numerical values, to format the paper to meet desired properties, such as a defined page limit [9].…”