Multimedia BEND: Augmenting Bend Analytics with Visual and Audio Cues

Our objective is to design, develop and test the next generation of BEND analytics accounts for visual and audio cues. Grounded in psychology, sociology and cognitive science, we propose a principled approach that extracts from visual/audio channels both emotional cues for each of the primary emotions (e.g., excitement and anger), and community cues for common identities (e.g., family, friend, group).  Visual, audio, and existing semantic cues will be used to estimate which BEND maneuvers are present.  Reinforcing multimedia cues will increase the likelihood of the maneuver, and contradictory cues may indicate sarcasm or humor and thus the distortion or distraction BEND maneuvers. The focus on primary emotions and identities enables cross-cultural relevance and cross-platform applicability which will be tested using data from diverse regions and social media. The result of this effort will be new theoretically grounded scalable multimedia algorithms integrated with the existing BEND maneuver software, and increased accuracy in influence campaign assessment. Our second objective is to develop illustrative use cases and operational guidelines to support the warfighter in using these analytics. Meeting both objectives will support meaning making and decision making regarding influence, course of action assessment and forecasting of behaviors and events.