Inference of Human Behavior via Network Mining

BEING was a project from COPELABS dedicated to exploratory research activities in the context of mobile crowd sensing. In BEING we rely on middleware developed in the COPELABS CitySense (2012-2017) project and focus on contextualization derived from data captured via wireless networks, and which today any device overhears. BEING expects to contribute to the development of pervasive wireless solutions, as well as studies in mining in wireless networks, to assist in a better and non-intrusive inference of human behavior, from an individual and collective perspective.



BEING considers behaviour inference to be a product of three main aspects: I) passive sensing (roaming context); ii) routine and activities (human context); iii) and contextual information (affinities and similarities). Such behavior inference is to be performed with the owners of sensorial devices being agnostic and not disturbed with any sensing activities, and focusing on adequate activity classification modeling, considering models which can incorporate sensing fusion (more than one sensor and the data correlation towards the prediction of e.g., activities).

In order to target the most important challenges of devising large-scale personal sensing systems, this project aims to investigate and develop a pervasive wireless sensing framework able of leveraging cooperative sensing, data and context gathering. In the proposed framework, only people with strong similarities share training data: a wide range of affinity networks (e.g., proximity networks, online social networks, conversation networks) indicates different typesof interpersonal similarities. The proposedframework will make use of these different networksto determine an affinity graph that quantifies the similarity betweenusers, identifying opportunitiesfor effective sharing.