Personal Context Recognition via Reliable Human-Machine Collaboration

An effective context recognition system cannot rely only on sensor data but requires the user to collaborate with the system in providing her own knowledge. In approaches such as participatory sensing, which leverages on users to annotate and collect their own data, user-generated data is usually assumed to be accurate; however, in real life situations, this is not the case. Research in social sciences and psychology shows that humans are unreliable due to several behavioral biases when describing their everyday life. In this paper, we propose to parametrize two biases, i.e., cognitive bias and carelessness, in order to identify and evaluate their impact on the users' reliability when recognizing users' context. The parameters are part of an architecture for context modelling and recognition from previous work, which combines sensors and users as a source of information. We evaluate our approach on a dataset of location points from the SmartUnitn One experiment.