[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
AMIDEN is the architecture to provide various services by making appliances' functions cooperate on home network. This architecture has a problem that user's load of selecting functions increases with increasing number of appliances on a home network. To deal with this problem, we propose a framework for function recommendation based on user models of tendency of usage. Our framework doesn't force any change in the process which controls each appliance. This makes it possible to reduce the cost of introducing our framework to AMIDEN enabled environment. In our framework, user model is pluggable so that function recommendation system can adopt appropriate model for its purpose. And, function recommending process runs in parallel with service construction process. This makes it possible to suppress increase of service construction time. Moreover, time threshold is set for communication with recommendation server so that service construction can be accomplished without the influence of server down.
We implemented this framework on ``Ukari Core'', a prototype implementation of AMIDEN. First, we experiment in a virtual home network which consists of 30 virtual appliances, by using a service which present a picture image made by a virtual digital camera. The result of this experiment shows that the framework can make appropriate function recommendation according to given user model. And we confirm that the average increase of service constructing time is suppressed within less than 2 seconds, by measuring service constructing time before and after introducing our framework. Moreover we confirm that service construction is accomplished even if the recommendation server is down. Secondly, we experiment in a home network which consists of 4 real appliances, by using two services; a service which presents a image of a person detected by a sensor, and a service which presents a image of inside of a refrigerator. Through the experiment, we confirm that history data is normally recorded on our framework.