| Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 94 === In this thesis, a dynamic personal preference modeling system is proposed. The system is able to learn the user''s preference model and adjust this model according to some changes(the user''s behavior, the environment, etc.) in a Smart Home, and hence the system can provide adequate services to the inhabitant.
First, we need a data collection mechanism to collect all the information in a Smart Home, and there are two kinds of data which have close relationships with the personal preference model: the environment data and the personal predefined data.
The First kind consists of all pure data gathered from the environment, including sensory information, electric appliance state, personal information, time information, etc. The other kind consists of personal predefined data, i.e., data predefined by the user, which can help the system to adjust the personal preference model automatically.
After collecting these two kinds of data, the system uses Bayesian network with semi-supervised learning to build the personal preference model and predicts the personal preference via this model. Our system can provide service to the user and infer whether the user''s behavior changes or not. Then, the system can interact with the user to obtain some useful information and adjust the preference model according to such information.
|
|---|
