Automatic Estimation of Topic Flow from Slides for Supporting Presentation Slide Creation

概要

論文の詳細を見る
Researchers often present their ideas using presentation tools such as Microsoft Office PowerPoint. However, listeners sometimes cannot follow the story and cannot understand what the speakers intend, even if topics are appropriately selected and arranged. Such situations are caused by inappropriate representation of sentences that describe topics. The audience cannot re-organize topic structures based on the descriptions on the slides. Our research is aimed at developing a system that automatically detects differences in the speaker's (or author's) intention for topics and a topic structure that can be estimated from created slides. We focus on a mechanism for automatic estimation of the relationships between slides, especially sequential and inclusive relationships. Relationships between slides can be inferred by the change in focus in the topic structure, and the topic structure is estimated using lexical information on the slides. Therefore, we introduce a topic graph that represents the relationships among topics on slides, which is formed using lexical and layout information of slides. Relationships between created slides are detected by the change in their focused topics in the topic graph. Based on an experiment, we validated the detected relationships with a prototype system when slides mainly consist of typical topics of a presentation and do not contain compound nouns.