E-science is moving from grids to clouds. Getting the best of both worlds needs to build on the experience gained by the steady operation of production grids since some years. With the Grid Observatory initiative, trace data are publicly available to the computer science and engineering community and can be used for dimensioning and optimizing infrastructure. This paper proposes a new approach for analyzing behavioral traces: as most of them are indeed text documents, state of the art techniques in text mining, and specifically Latent Dirichlet Allocation, can be exploited. The advantages are twofold: providing some level of explanation inferred from the data; and a relatively scalable way to capture the temporal variability of the behavior of interest, while retaining the full dimensionality of the problem at hand. We experiment the text mining analogy approach by characterizing file access behavior. We validate the resulting probabilistic model by showing that it is capable of generating synthetic traces statistically consistent with the real ones.