Identification of Events in Use Cases

Context: Requirements are one of the most important elements of software development. They involve many stakeholders, including IT professionals (Analysts, Project Manager, Architects, etc.) and IT laymen (e.g. Customers, End Users, Domain Experts, Investors). In order to build a bridge between those two groups, requirements are often written in a natural language, e.g. in the form of use cases. They describe interaction between an end-user (called “actor”) and the system being built. The most important element of a use case is a main scenario leading to obtaining actor’s goal in the most “typical" way. A main scenario is a sequence of steps taken by the actor or the system. Each step describes a simple action in a natural language (e.g. “Student enters his account and password."). Deviations from the main scenario are also documented within a use case. They are usually called extensions and are described as alternative interactions. Each extension is a response to a given event causing the deviation (e.g. “Account or password is wrong."). The first problem is effectiveness of event identification, i.e. degree of completeness of identification of the events causing deviations from the main scenario (in an ideal case all the events should be identified). Another problem is efficiency, i.e. time necessary to identify events. Identifying all such events is very important, as unidentified events could negatively impact not only coding but also effort estimation, user interface design, user manual writing, architecture evaluation, etc. In general, three approaches to event identification are possible: 1) an ad hoc approach used by humans, 2) a specialised method of event identification to be used by humans, 3) an automatic method to be executed by a computer. Objective: The aim of the thesis is to propose an automatic method of event identification that would not be worse than identification made by humans, in terms of effectiveness (event completeness) and efficiency (time). Method: To develop an automatic method of event identification, a pool of almost 200 use cases was divided equally into a learning part and a testing part. Through analysis of the learning part, 14 abstract event types and two inference rules were identified. Using NLP tools (Stanford Parser, OpenNLP) and a use case editor (UC Workbench), a pilot version of a tool for automatic event identification in use cases was designed and implemented. The tool was used to evaluate both the effectiveness (event completeness) and efficiency (time) of the proposed method against results obtained from human-based experiments performed in academia (18 students) and industry (64 practitioners). 9 students and 32 practitioners used the ad hoc approach, similarly 9 students and 32 practitioners used a HAZOP-based method called H4U. Both the human-based experiment and the evaluation of the proposed method were based on a use-case benchmark developed at the Poznan University of Technology, Faculty of Computing. Results: From the thesis it follows that: • the average effectiveness of event identification performed by humans using the ad hoc approach is at the level of 0.18; • the average effectiveness of event identification performed by humans using the HAZOP-based approach is at the level of 0.26 and the difference between those two approaches is statistically significant; • an automatic tool for event identification can be created and its average effectiveness is at the level of 0.8 (the automatic method outperformed humans and this is statistically significant); • the speed of analysis of use cases, for the purpose of event identification, with the ad hoc approach is at the level of 2.5 steps per minute and with the HAZOP-based method is at the level of 0.5 steps per minute; • the speed of analysis of use cases by the automated tool, for the purpose of event identification, is at the level of 10.8 steps per minute; • using NLP tools one can automatically identify 9 bad smells in use cases which are based on the good practices proposed Adolph et al. in their book Patterns for Effective Use Cases. Conclusions: It is possible to build a commercial tool aimed at automatic event identification in use cases, that would be a part of an intelligent use-case editor. Such a tool would support breath-first approach to elicitation of use cases and thus would fit the agile approach to software development. Politechnika Poznańska Wydział Informatyki Identyfikacja zdarzeń w przypadkach użycia