Question Answering communities have known a large expansion over the last few years. Reliable people sharing their knowledge are not that numerous. Thus, detecting experts since their first contributions can be considered as a challenge. We are interested in studying the activity of these platforms’ users during a defined period of time. As the data collected is not always reliable, imperfections can occur. In order to manage these imperfections, we choose to use the mathematical background offered by the theory of belief functions. People say that the more time they spend within a community, the more knowledge they acquire. We investigate this assumption in this paper by studying the behavior of users without taking into consideration the reputation system proposed by Stack Over ow. Experiments with real data from Stack Over ow demonstrate that this model can be applied to any expertise detection problem. Moreover, it allows to identify potential future experts. The analysis allows us to study the behavior of experts and non expert users over time spent in the community. We can see that some users keep on being reliable while others do gain knowledge and improve their expertise measure.