Frame selection using iterative grammatical evolution for automatic comic generation from game logs

Recently, we developed a system that visualizes a game play, from its game log, with a comic-style representation. In that system, comic frames are selected based on their interaction levels without considering users´ interests. In this paper, we tackle this issue by proposing an interactive frame-selection method that adopts interactive evolutionary computation (IEC) for evolving conditional rules used in selection of frames. IEC enables the user to directly give feedbacks on which an originally proposed subjective fitness function bases. In order to lessen user fatigue, the proposed method evolves the frame-selection conditional rules for certain consecutive generations based on this subjective fitness function - taking into account the latest user feedback - and another originally proposed objective fitness function. The objective fitness function ensures the consistency in the ratio among event types in a resulting comic with the ratio in the game log. As a promising evolutionary computation technique for the task in this work, we use grammatical evolution. We then evaluate the proposed method by a user evaluation using typical game logs. Experimental results confirm that evaluation participants favor comics whose frames are selected by the proposed method more than comics generated by a baseline method.