PlotThread: Creating Expressive Storyline Visualizations using Reinforcement Learning

摘要: Storyline visualizations are an effective means to present the evolution of plots and reveal scenic interactions among characters. However, design storyline is a difficult task as users need balance between aesthetic goals narrative constraints. Despite that optimization-based methods have been improved significantly in terms producing legible layouts, existing (semi-) automatic still limited regarding 1) efficient exploration space 2) flexible customization layouts. In this work, we propose reinforcement learning framework train AI agent assists exploring efficiently generating well-optimized storylines. Based on framework, introduce PlotThread, authoring tool integrates set support easy visualizations. To seamlessly integrate into process, employ mixed-initiative approach where both designers work same canvas boost collaborative We evaluate model through qualitative quantitative experiments demonstrate usage PlotThread using collection use cases.