Machine Learning and Constraint Solving for Automated Form Testing

Abstract

In recent years there has been a focus on the automatic generation of test cases using white box testing techniques, however the same cannot be said for the generation of test cases at the system-level from natural language system requirements. Some of the white-box techniques include: the use of constraint solvers for the automatic generation of test inputs at the white box level; the use of control flow graphs generated from code; and the use of path generation and symbolic execution to generate test inputs and test for path feasibility. Techniques such as boundary value analysis (BVA) may also be used for generating stronger test suites. However, for black box testing we rely on specifications or implicit requirements and spend considerable time and effort designing and executing test cases. This paper presents an approach that leverages natural language processing and machine learning techniques to capture black box system behavior in the form of constraints. Constraint solvers are then used to generate test cases using BVA and equivalence class partitioning. We also conduct a proof of concept that applies this approach to a simplified task management application and an enterprise job recruiting application.