A semantic-based biosignal mining for Affective Computing

Affective computing is the research field in charge of the computational processing of emotions. Applications range from enhancing human-computer interaction to biofeedback generation for avoiding risky situations and improving subjects' well-being. Current automatic emotion recognizers basically provide only simplistic classication approaches and yield trivial labels using machine learning techniques without capturing relations between biosignals and observations measured by the various sensors. This thesis proposes a framework for affect recognition monitoring human body vital signals through wearable, non-invasive sensors. The intrinsic complexity in emotion detection is tackled by means of automatic discovery based on a semantic matchmaking process via non-standard reasoning services. In particular, manipulation of vague concepts such as emotions and their dynamic evolution are achieved exploiting Fuzzy Description Logics (DLs) ontology-based approach. The proposed architecture is able to receive in input time series of biosignals, extract meaningful high-level knowledge from mining, identify emotional patterns through the flexibility of Fuzzy-DLs, and exploit semantic-based matchmaking to recognize user emotions. Prototypes were implemented w.r.t. a reference dataset and preliminary experimental tests were carried out to verify the feasibility of the approach on emotions experienced by users. The system enhances human-computer interaction allowing a feedback generation for subjects and improvement of their well-being. Motivations for the work stem from the idea that a revision of existing approaches can have a signicant impact on the effectiveness and applicability of affective computing, by combining logic-based Knowledge Representation (KR) with machine learning techniques in a low-cost wearable computing set-up. The social aspects of improved Affective Computing systems concern mainly the potential applications in diagnosis, treatment and management of mental and stress-related disorders. The possibility to support biofeedback in users to oppose undesirable emotions and behaviors can also have a significant impact on substance abuse and other unhealthy habits, affecting the quality of life of the general population and welfare policies.