ANALYSIS OF USER TRAJECTORIES BASED ON DATA DISTRIBUTION AND STATE TRANSITION: A CASE STUDY WITH A MASSIVELY MULTIPLAYER ONLINE GAME ANGEL LOVE ONLINE

At present, trajectory data, series of coordinate data traversed by moving objects, can be readily obtained due to the advent of positioning technologies. Clustering of trajectories and giving meanings to the resulting clusters is an active research area. Recently, we proposed an analysis approach that clusters trajectories in two steps: the first step based on data distribution and the second step based on state transition. In this approach, for coping with the distinguished characteristic of each trajectory, a map of interest is dynamically divided into multiple states, according to the trajectory distribution, and a quadtree is generated for each trajectory. The first-step clustering is then performed based on the differences between the quadtrees. For all trajectories in a resulting cluster of interest, the second-step clustering is further performed based on the differences in their state-to-state transition probabilities using a proposed method for comparing a pair of trajectories with different quadtree structures. In this paper, after presenting a procedure for visualizing a cluster of interest in order to interpret its movement behaviors, we give and discuss a case study where our approach is applied to real trajectory data obtained from Angel Love Online, a massively multiplayer online game, manifesting player behaviors in the target map.