Identificação automática de problemas encontrados e mudanças sugeridas por usuários em seus comentários sobre aplicações móveis

Abstract

The popularization of mobile devices and the increase of its presence in people’s daily lives has induced the development of countless software systems, build specifically for such platform and known as mobile applications, or apps for short, created with the intent of adding value to its users’ lives, by offering many kinds of funtionalities and services. Currently, the app stores centralize many operations related to apps, and one of them is allowing users to review and rate the application. Reviews can function as a source of direct feedback to the developers about various aspects that refer to ther apps, that vary between compliments, criticisms, suggestions, and so on. Among the information the users provide, stand out those that mention problems found and suggestions of possible changes or improvements to the application, which are able of directly helping the developers to evolve their software, since they make explicit or emphasize specific features it presents, or that could be added. However, though it is possible to utilize the content of these reviews to increment and improve the app, it is needed to pinpoint which of them contain said useful information, and which topics the users quote. Such process would demand great effor and time, if executed manually, mainly for more popular applications. In this context, the current work proposes the use of clustering, a data mining technique that realizes grouping of objects based on how similar they are with each other, to try gathering reviews that talk about the same themes or aspects app related. Thereunto, the clustering process was applied to user reviews over the Facebook and Telegram apps, in their versions for the Android operating system, taking into account two distinct similary measures, with the purpose of analyzing how it may help the automatic identification of problem reports e change requests. The results show that there are groups very cohesive in respect to the themes in its reviews, allowing the derivation of kinds or classes of problems from the cluster’s description. Moreover, there are indications that the presence of problems and the absence of certain features impact the ratings given to the app. Lastly, it was observed that the distances taken into account produce distinct results but, in general, Jaccard generates more cohesive clusters regarding tópicos within them.