Abstract
Mining object movement patterns to understand the behavior of moving objects has many high impact applications. However even if existing pattern models are meaningful, there are still challenging issues such as: 1) there are many kinds of patterns and algorithms without an efficient management scheme, 2) lacking of relevant approaches in dealing with time gaps, 3) they usually focus on an unchanged group of objects and thus do not capture the behavior of the whole object class, 4) Naturally, there are huge amount of very redun- dant movement patterns extracted while only a few of them are meaningful. However, few researchers address this issue. 5) Despite the growing demands for diverse applications, there have been few scalable tools for mining massive and sophisticated moving object data.In my thesis, I will mainly focus on addressing these issues. We propose the three step framework: 1) the first step aims to present and framework to mine and manage different existing movement patterns in an efficient way, 2) in the second step, we propose novel movement pattern concepts to access the relevance of movement patterns by dealing with time gaps. We also further present the gradual trajectory pattern notion to analysis the behavior of moving objects in a graduality point of view. 3) In the last step, we propose an novel MDL principal-based approach, named SmartCompo to extract representative movement patterns from moving object data. 4) Our three step framework is illustrated in a demonstration system, named Multi_Move, which is designed to extract and manage dif- ferent kinds of spatio-temporal patterns concurrently. A user-friendly interface is provided to facilitate interactive exploration of mining results.As an extra work, I further present the concept of multi-relational gradual pattern which generalizes the gradual pattern notion in single relation data to multi-relation database.