Humans implicitly understand social connections involving character roles in a storyline, either by reading a novel, or watching a movie. We present a project for bridging the social graphs obtained from verbal stories, such as a novel, with the social connections extracted directly from a video. We present the system architecture, and outline the current progress, including extracting main characters from videos, and building story contexts to represent character presence patterns, in order to match with time-varying social graphs. We study several direct benefits of this approach, and expect the individual graphs extracted can be combined toward building a social knowledge graph covering a broader context and a range of character roles. This project is being conducted in Samsung Research AI Center, and utilizes Bixby services for communications between the NLU backend, the Characters database, and an interactive frontend. 

https://www.kdd.org/kdd2019/docs/KDD2019_Showcase_2056.pdf