ConaMSN: مسنجر آگاه از متن با استفاده از شبکه های بیزی پویا با سنسورهای پوشیدنی

With the growth on the concern about context-aware applications, it becomes important to recognize and share user context. Even though there are some applications, it is still limited in managing simple contexts. In this paper, we propose a context-aware messenger application that exploits dynamic Bayesian networks to automatically infer a user’s context and shares contextual information to enrich electronic communication. It collects various sensory information and displays representative user contexts such as emotion, stress, and activity in the form of icons in the messenger program.

People want to enrich their social relationship by communicating with each other, and implicit situational information, called context, is beneficial to increase the richness of communication (Griswold et al., 2004 and Ranganathan et al., 2002). However, it is hard to share the context of users who are far apart from each other. Recently, a messenger (like Windows Live Messenger) for chatting with someone electronically provides a function to show a simple context such as online or offline, in business or in break, etc. Even though these kinds of information might be useful to properly contact with others, it is still limited to share the wide variety of contexts that could be sensed in a face-to-face conversation. We thus aim to create a system that could recognize various contexts such as the level of stress, the type of emotion and activity. Our goal is to recognize a user’s context based on information collected from wearable sensors and to share the context through a messenger application. Our system, context-aware messenger (ConaMSN), manipulates the large number of raw data to enlarge the simplicity of conventional context-sharing systems. From a user’s physiological information and movement collected by using Armband and accelerometers, ConaMSN infers various contexts with modular dynamic Bayesian networks (BNs) and visualizes them with a set of icons. By exchanging contextual information of users, ConaMSN lets them know their buddies’ situation and improve electronic communication. For example, users can give words of encouragement to their friends in a gloomy mood. Also, they can expect a slow response, since they already know that their buddies are involving in some activity.

In this paper, we developed a context-aware messenger application which inters three types of contexts such as emotion, stress, and activity from sensory signals by using probabilistic models, and verified the usefulness of the proposed system with several experiments. Our long-term goals are to evaluate the system using large-scale real logs collected over a long time period from various subjects and to extend the variety of contexts for sharing. It might be useful to tune and evaluate the inference model of BNs and DBNs with the data collected. The proposed system still has some weak points. For some contexts such as emotion and stress, it is not easy to label samples correctly and to construct the inference model. It is sensitive to the variation between different people, so we need to develop a more sophisticated learning algorithm for personalized context recognition.