Using Natural Language Processing Techniques to Improve Manual Test Case Descriptions

Despite the recent advancements in test automation, testing often remains a manual, and costly, activity in many industries. Manual test cases, often described only in natural language, consist of one or more test steps, which are instructions that must be performed to achieve the testing objective. Having different employees specifying test cases might result in redundant, unclear, or incomplete test cases. Manually reviewing and validating newly-specified test cases is time-consuming and becomes impractical in a scenario with a large test suite. Therefore, in this paper, we propose an automated framework to automatically analyze test cases that are specified in natural language and provide actionable recommendations on how to improve the test cases. Our framework consists of configurable components and modules for analysis, which are capable of recommending improvements to the following: (1) the terminology of a new test case through language modeling, (2) potentially missing test steps for a new test case through frequent itemset and association rule mining, and (3) recommendation of similar test cases that already exist in the test suite through text embedding and clustering. We thoroughly evaluated the three modules on data from our industry partner. Our framework can provide actionable recommendations, which is an important challenge given the widespread occurrence of test cases that are described only in natural language in the software industry (in particular, the game industry).