کاربرد نظریه های خطای انسانی برای بهبود فرایند مهندسی مورد نیاز

Problems with the structuring and execution of the Requirements Engineering (RE) process can lead to project failure. As the involved activities are essentially human and subjective, the importance of the human factor must be considered. The possible correlation between human errors and the problems that may occur in the RE process is explored in this paper in order to propose a process improvement approach, which applies the theories of human error to build an expert system for automated diagnosis of problems. This approach starts by assessing the importance of individual types of human errors in different contexts using questionnaires to extract knowledge from RE practitioners. A method for managing the priorities of solutions to improve a given process is then proposed and a prototype application is implemented for evaluation with practitioners. The activities with RE practitioners also provide data that are used to construct and evaluate estimators of the presence of human errors and the affected phases of the RE process. Most of the volunteering participants agree that knowledge of error typologies may help to improve their processes to avoid problems.

The purpose of this research is the tentative improvement of Requirements Engineering (RE) processes. Problems during the activities of the RE process, such as imprecise plans, loss of information and information recorded in ambiguous or incomplete form [8] can cause projects to fail, delay, have the products delivered with deficient quality and overrun the stipulated budgets. Some examples of problematic projects have been reported to us, including cases of a team trying to specify a system for about 2 years without being able to release a single version for production, teams reporting lack of access to final users, managers with disparate conflicting goals and lack of people with domain knowledge for a specific project. As most of the activities of RE are essentially human and subjective and, therefore, susceptible to human error, the Human Error Theories (HETs) may assist the understanding of the problems in the process, and the adoption of controls and methods for their prevention and detection. By enhancing perception of the problems and pointing to the estimated appropriate solutions, our research may help to solve the aforementioned kinds of problems. The supposed correlation between human error and problems in the RE process is analyzed in this work in an exploratory and quantitative manner, by proposing and evaluating forms of application of the HETs in RE. The principal proposed form of application considers the construction of an expert system to produce an ordering of human error types by importance for specific activities of RE. This order by importance is obtained at first in a global context and later refined for particular contexts of the adopted processes. The refinement of the importance of the error types is then used to select known solutions that must be implemented according to their priorities so as to improve the process. Thus, the proposed method can be seen as a solution management device to allow the incremental improvement of an already adopted process. The methodological approach employed in this paper is based on knowledge extraction from RE practitioners. Questionnaires were elaborated for RE practitioners to help to evaluate the importance of human errors and to identify particular types of error with greater importance for selected activities of RE. For this, two activities were selected as they suit well the intent of the investigation due to having a predominance of human intellectual work that is hard to automate. The types of human errors were obtained primarily from the theories, which include their classification, dynamics and detection, prevention and mitigation techniques. With the information obtained from the questionnaires, we proposed and evaluated a method for solution management based on process diagnosis. For this, we assumed that the importance of these error types is dependent on the adopted process and that the improvement of the process may be guided towards the implementation of solutions related to error types of greatest importance. A simplified diagnosis method was developed in order to estimate the relation between selected problems and the possible involved human error types. The employed list of problems and solution suggestions was compiled from the literature on RE and human error. To evaluate the acceptance of the method, an application prototype was produced and applied as part of an activity with volunteering RE practitioners. From the interviews carried out with the group of volunteering RE practitioners and their consequent interpretation, it was possible to gain greater understanding of the influence of human error in the RE process. We noticed in the responses: (1) the occurrence of error types with high importance for each distinct activity, (2) the possibility of existing latent error types, (3) the varied degree of importance of error types for different contexts, (4) a possible relationship between the perception of the importance of the error type and the stage in the human action cycle in which it took place, (5) the perception of the RE practitioners of the need for a tool to help organize the activities of the process, and (6) the awareness of how to identify and treat the human error in their activities. In addition, based on data acquired from RE practitioners, classifiers able to estimate the presence of error types and the most affected phases of the process were constructed. This paper is organized as follows. In Section 2, a review of related work considering both the aspects of human error and the improvement of the RE process is presented. In Section 3, the most relevant points of the HETs are discussed: the definition, the nature, the typology and the dynamics of the human error. In Section 4, the conceptual RE process adopted in this work is presented. Section 5 describes the research methodology: the method for knowledge acquisition from the RE practitioners through questionnaires as well as the method of diagnosis and solution management, including ways to evaluate them. Section 6 reports and discusses the results of the many parts of the adopted methodology. Section 7 finishes the paper, presenting the conclusions and a summary of hypotheses to investigate in the future.

This work sought to bring the standpoint of the HETs to the process of RE in order to improve the process by preventing or minimizing the human error in it. First of all, the importance of the many types of human errors was assessed by analysis of the responses collected from practitioners of RE through questionnaires. Some results of this analysis are displayed as dispersion maps of the elicited types of error according to frequency and criticality for two selected activities of RE, namely, stakeholder interaction and requirements description. This type of plot allows us to visually detect the most important types of human error for the RE process or specific activities of this process. The resulting method to assess the importance of human error may be systematically applied with the participants of a process to estimate the most important error types of specific contexts or particular activities. In addition, we proposed and evaluated a method for solution management applied to RE. This proposed method lists, ordering by importance, the types of human error that may occur given the list of present problems in a process. To do this, the method depends on process diagnosis and was developed to have an adaptive nature so it can be implemented over an existing process by selecting, in a lean manner, only solutions of more immediate importance. The diagnosis provides a list of the error types related to the set of process problems with an importance measure, and corresponding well-known solution alternatives from the literature are suggested. For the evaluation of the method, a prototype system was implemented. The testimonies of the RE practitioners that participated in the evaluation revealed the didactic aspect of the tool. The prototype was well-received by the respondents, who considered the tool to be relevant and coherent and were convinced that the awareness of human error, its typology and prevention may contribute to improve RE processes. The proposed diagnosis method for solution management is based on classification techniques. Three different techniques were evaluated by training the classifiers with data collected from RE practitioners using a questionnaire and applying the cross-validation technique to enforce the generalization. Although the origin of the data was very subjective and imprecise, and it was hard to find volunteers to provide a bigger sample, some accurate discriminations of some of the error types and phases of the process were obtained. The identification of the occurrence of error types and most affected phases from the set of perceived problems, as long as good accuracy is provided, satisfies the need for a diagnosis system to point out both the error types and the process phase in which they occur. These data were also analyzed using a clustering method. The clustering method produced better results for error type identification and process phase identification. It may imply that instances can be grouped in cases with similar problems. Such consideration may facilitate the analysis of problems in the process and the proposal of solutions. For future work, we intend to investigate some open hypotheses that can help to benefit the process of RE. For example, we noticed that the perceived importance of errors may depend on the related stage of the cognitive process. The practitioners tended to consider errors in the perception stage more important than errors in the action stage. If this is a rule, then the selection of problems should be organized according to the cycle of human information processing. Another example is the identification of latent errors, which rarely occur, but may have a great impact. Such analysis may allow the detection of problems even in processes considered generally good, but failing to deal with unexpected events. There is also opportunity for improvement of the process diagnosis and solution management tool, which should be evaluated in practical circumstances. The performance results of the diagnosis with clustering suggest that case-based reasoning might be a fruitful model to investigate. Additional studies concerning the set of observed variables are suggested. For example, errors can be grouped in fewer categories such as considering a single category for lapses instead of 5, or some error types that may not be relevant to accomplish discrimination could be excluded. Some additional variables that are more objective and directly observable should be studied to improve classifier performance. Finally, greater attention should be given to important concerns of RE such as problems related specifically to nonfunctional requirements, domain knowledge and the requirements oriented towards either goals or means.