• Stigmergy
• Swarm Intelligence
• Spatio-Temporal patterns
• Pattern Analysis

Large amount of data are currently produced by an incredible diversity of applications. In order to provide relevant information to the users is necessary to identify important patterns and recognize these patterns when they occur again. In general, at the core of the analysis of human centric data is the construction of two possible types of model: (i) knowledge based models, explicitly designed at the business level in terms of logical or mathematical rules, determined by a domain expert; (ii) data-driven models, i.e., systems that can learn from prototypical data via machine learning or statistical algorithm. Nevertheless, modeling and reusing application contexts remains a difficult task. An important lesson learned is that the algorithms performing the parametric data aggregation must use a limited number of states, be highly adaptable and handle variability. The data-driven approach discussed in this work takes inspiration from the emergent paradigm, in which context information is augmented with locally encapsulated structure and behavior. Emergent paradigms are based on the principle of self-organization of data, which means that a functional structure appears and stays spontaneous at runtime when local dynamism in data occurs.
More precisely, we adopt marker-based stigmergy, i.e., a biologically inspired mechanism performing scalar and temporal information aggregation. In biology, stigmergy is an indirect communication mechanism, while in computer science stigmergy can be employed as a dynamic, agglomerative, computing paradigm because it embodies the time domain. Stigmergy focuses on the low level processing, where individual samples are augmented with micro-structure and micro-behavior, to enable self-aggregation in the environment.
We present the use of stigmergy computation in different applications with spatio-temporal data, showing the feasibility and the capability of the approach to be adopted in heterogeneous fields. To support the user in the parameterization process, we designed an adaptation mechanism based on a bio-inspired evolutionary algorithm. At the final stage of the architecture development, we compared the proposed approach with state-of-art techniques on a classification task. Experimental studies completed for real-world data show that results are promising and consistent with human analysis.