Scene Linking Annotation and Automatic Scene Characterization in TV Series
Abstract
In the context of a large collection of multimedia documents, creating links between documents or scenes can help to organize the collection. For TV series, this organization can be achieved by means of narrative structure extraction through scene linking. Narrative characteristics such as speaking characters, entity mentions and theme can be used to characterize scenes. The linking of scenes can be between scenes inside an episode, between scenes in di↵erent episodes and/or in di↵erent seasons, since stories in TV series progress at di↵erent level of granularity. In this work, we have annotated the links between the scenes of the first two seasons of the TV series Game of Thrones, using predefined stories and sub stories. We have also automatically extracted the narrative characteristics of each scene. The dataset is composed by 444 scenes, involving 154 speaking character organized in 46 stories divided into 151 sub stories and 5 sub sub-stories.
Origin : Files produced by the author(s)