Multimodal Sentiment-Evolution Analysis for Latent-Risk Estimation

The massive influx of information through the web and its rapid expansion through social media provides a rare but effective sliver of operational information. One core challenge in the extraction of operational intelligence is the share size of the data and the large level of noise (e.g., miss-information and fake news). It is critical that public social media and web information is efficiently analyzed to produce actionable intelligence. This intelligence can be used to help protect the nation's sovereignty and increase the safety and security of the US forces (soldiers, sailors, airmen, and marines operating in a coalition environment) with regards to inside and outside threats. The proposed sentiment-evolution and latent-risk estimation and analysis platform leverages powerful techniques in sentiment analysis for unimodal data (text, audio, visual) and combines their benefits first into a multimedia sentiment analysis module and second into a multimodal sentiment analysis module. The multimodal analysis includes a multimodal spatio-temporal data correlation task. The latent-risk estimation module uses sentiment-volution and real, historical data as a case study. Phase II of this proposal will extend the sentiment direct and indirect manipulation using information from social-media and socio-economic sources.