International audienceLarge amounts of traces can be collected by Pervasive Information Systems, reflecting user's actions and the context in which these actions have been performed (location, date, time, network connection, etc.). This article proposes refinement strategies with different frequency measurements on contextual elements in order to better analyze the impact of these elements on the user's behavior. These strategies are based on data mining and Formal Concept Analysis and used to refine input data in order to identify the context elements that have a strong impact on user behaviors. We go further on context analysis by cognizing FCA with semantic distance measures calculated based on a context ontology. The proposed context analysis is further on evaluated in experiments with real data. The novelties of this work lies on these refinement strategies which can lead to a better understanding of context impact. Such understanding represents an important step towards personalization and recommendation features

-This paper presents a new method for automatically extracting smartphone users' contextual behaviors from the digital traces collected during their interactions with their devices. Our goal is in particular to understand the impact of users' context (e.g., location, time, environment, etc.) on the applications they run on their smartphones. We propose a methodology to analyze digital traces and to automatically identify the significant information that characterizes users' behaviors. In earlier work, we have used Formal Concept Analysis and Galois lattices to extract relevant knowledge from heterogeneous and complex contextual data; however, the interpretation of the obtained Galois lattices was performed manually. In this article, we aim at automating this interpretation process, through the provision of original metrics. Therefore our methodology returns relevant information without requiring any expertise in data analysis. We illustrate our contribution on real data collected from volunteer users.

International audienceSocial content generated by users' interactions in social networks is a knowledge source that may enhance users' profiles modeling, by providing information on their activities and interests over time. The aim of this article is to propose several original strategies for modeling profiles of social networks' users , taking into account social information and its temporal evolution. We illustrate our approach on the Twitter network. We distinguish interactive and thematic temporal profiles and we study profiles' similarities by applying various clustering algorithms, by giving a special attention to overlapping clusters. We compare the different types of profiles obtained and show how they can be relevant for the recommendation of hashtags and users to follow