Is hybrid AI suited for hybrid threats? Insights from social media analysis

Social media create the opportunity for a truly connected world and change the way people communicate, exchange ideas and organize themselves into virtual communities. Both understanding online behavior and processing online content are of strategic importance for security applications. However, high volumes, noisy data and rapid changes of topics impose challenges that hinder the efficacy of classification models and the relevance of semantic models. This paper performs a comparative analysis on supervised, unsupervised and semantic-driven approaches used to analyze social data streams. The goal of the paper is to determine whether empirical findings support the enhancement of decision support and pattern recognition applications. The paper reports on research that has used various approaches to identify hidden patterns in social data collections where text is highly unstructured, comes with a mix of modalities and has potentially incorrect spatial-temporal stamps. The conclusion reports that the disconnected use of machine learning models and semantic-driven approaches in mining social media data has several weaknesses.