Game Outlier Behavior Detection System Based on Dynamic Time Warp Algorithm

Abstract

This paper proposes a methodology for using multi-modal data in gameplay to detect outlier behavior. The proposed methodology collects, synchronizes, and quantifies time-series data from webcams, mouses, and keyboards. Facial expressions are varied on a one-dimensional pleasure axis, and changes in expression in the mouth and eye areas are detected separately. Furthermore, the keyboard and mouse input frequencies are tracked to determine the interaction intensity of users. Then, we apply a dynamic time warp algorithm to detect outlier behavior. The detected outlier behavior graph patterns were the play patterns that the game designer did not intend or play patterns that differed greatly from those of other users. These outlier patterns can provide game designers with feedback on the actual play experiences of users of the game. Our results can be applied to the game industry as game user experience analysis, enabling a quantitative evaluation of the excitement of a game.