André Di Thommazo

Using Information Visualization and Text Mining to Facilitate the Conduction of Systematic Literature Reviews

Systematic Literature Review (SLR or SR) and Systematic Mapping (SM) are scientific literature review techniques that follow well-defined stages, according to a protocol previously elaborated. Besides systematizing the search for relevant studies, the SR predicts the organization and the analysis of the obtained results. However, the SR application is laborious because there are many steps to be followed. Aiming to offer computational support to SR and SM, the StArt (State of the Art through Systematic Review) tool was developed. Besides helping the steps of SR or SM, the StArt tool has implemented visualization and text mining techniques to support the conduction and the reporting of the SR or SM. A comparative analysis was carried out in relation to StArt and other similar tools.

Requirements Traceability Matrix: Automatic Generation and Visualization

Background: Requirements management is considered one of the activities responsible for system failures. The difficulty regarding to requirements trace ability makes the system changes hard to be managed. Objective: This paper presents two approaches that allow the automated generation of the Requirements Trace ability Matrix (RTM): the RTM-E approach, which is based on the requirement input data, and the RTM-NLP approach, which is based on Natural Language Processing-NLP. Method: The RTM-E comprises the requirements dependency related to the data manipulated by them, while the RTM-NLP comprises the requirements dependency related to the similarities of their functionality descriptions. The results are shown through visualization of information in order to facilitate the understanding of such dependencies. Results: We conducted an experimental study in which both approaches were applied to 18 requirements documents. The RTMs created automatically were compared with the reference RTM created manually based on the stakeholders knowledge. Comparing the generated matrices, it was seen that the RTM-E on average matches 82% to the reference RTM, while the RTM-NLP approach on average matches 53%. Conclusions: The results show that generating the RTM based on these approaches, the effectiveness on determining the requirements dependences is satisfactory and motivates to keep studying in order to make improvements for both approaches.

Using the Dependence Level Among Requirements to Priorize the Regression Testing Set and Characterize the Complexity of Requirements Change

Background: When there are changes in software requirements, other phases of software development are impacted and frequently, extra effort is needed to adjust the previous developed artifacts to new features or changes. However, if the development team has the traceability of requirements, the extra effort could be not an issue. An example is the software quality team, which needs to define effective tests cycles in each software release. Goal: This papers aims to present an approach based on requirements dependence level to support the regression test prioritization and identify the real impact of requirement changes. Method: The designed approach is based on automatic definition of Requirements Traceability Matrix with three different dependence levels. Moreover, dependence between requirement and test case is also defined. A case study in a real software development industry environment was performed to assess the approach. Results: Identifying the dependence level among requirements have allowed the quality assurance team priorize regression tests and, by means of these tests, defects are early identified if compared with tests execution without priorization. Moreover, the requirements changes complexity is also identified with the approach support. Conclusion: Results shows that definition of dependence levels among requirements gives two contributions: (i) allowing test prioritization definition, which become regression test cycle more effective, (ii) allowing characterize impacts of requirements changes, which is commonly requested by stakeholders.

Using Visualization and Text Mining to Improve Qualitative Analysis

Context: Qualitative analysis is a scientific way to deeply understand qualitative data and to aid in its analysis. However, qualitative analysis is a laborious, time-consuming and subjective process. Aim: The authors propose the use of visualization and text mining to improve the qualitative analysis process. The objective of this paper is to explain how the use of visualization can support the Coding in multiple documents simultaneously, which may allow codes standardization thus making the process more efficient. Method: The Insight tool is being developed to make the proposal feasible and a feasibility study was performed to verify if the proposal offers benefits to the process and improves its results. Results: The study shows that the subjects who applied the proposal got more standardized codes and were more efficient than the ones who applied the process manually. Conclusions: The results derived from the use of visualization and text mining, even in a feasibility study, encourage proceeding with the project, which aims to combine both techniques to obtain more benefits on qualitative analysis conduction.