Information Technology Reference
In-Depth Information
the mood state of a user (based on the smartphones sensors' data), and generate
corresponding reactions to the detected mood. The methodology is described in
details by exploring mood recognition from smartphone sensory data, controlling
the behavior of the virtual companion and visualization of the working process.
The proposed model has been implemented and evaluated, which is described in
this paper.
2 Methodology
The proposed methodology is anchored on the idea of combining uncertainty
modeling with behavior networks, proposed in few previous works such as [3,4].
The improvement on the initial previously proposed ideas is made by considering
the mood history of user using DBNs, using more sensors from the smartphone
and fusion of inference techniques, as well as customized design of the behavior
network for the specific purpose of the artificial affective companion. The system
performs two functions of mood inference and behavior generation, which are
detailed below.
2.1 User Mood Recognition from Smartphone Data
The mood estimation module of the system (PME) is composed of SVM and
DBNs. SVM detects the physical activity level using accelerometer data and
DBNs is designed to infer and map the valence and arousal result to the two-
dimensional valence arousal model as illustrated in [8].
The DBNs is selected due to the uncertainties and dynamic environment of
the smartphone [7]. DBNs is well suited to the problem of inferring high-level
information such as the mood state by using the low-level data obtained from the
smartphone. That is because the mood state of the user depends probabilistically
on many different factors. For example, with the accelerometer, we can know
whether he is moving or not, and if so, whether he is moving fast or slow, and
with the GPS, we can estimate the user location, whether he is at home or at
work or somewhere outdoors. This low-level data obtained from the smartphone
sensors, can predict the conditional probability of the user status. By setting
the logic based, conditional probabilities of the user mood in relation to the
smartphone sensor data, the mood state could be estimated without the need of
precise knowledge and training the system such as in the method proposed by
[6]. Figure 1 shows the designed DBNs model. The model infers the valance and
arousal level of the user, which could be mapped to the Valance-Arousal Model
(VAM) and define the detected mood state.
2.2 Behavior Controller of the Virtual Companion
After detecting the users mood state, there is a need for an algorithm to control
the respective artificial companion behavior. In this respect, a behavior network
is designed. In the proposed behavior network, The environment and sensing part
