Analyzing Narrative Evolution Across Social Networks

Narratives organize and motivate social networks, enabling them to process and interpret complex events and take individual and collective action in response. However, while past work on narrative extraction and analysis provides useful ‘snapshots’ of a narrative at a given point in time, it generally fails to 1) distinguish over-arching, foundational narratives from concrete, focal events that impact those narratives, 2) track the evolution, merging, and splitting of narratives over time, quantifying changes caused by focal events, and 3) capture the interactive competition and coordination of distinct narratives as they spread within social networks and sub-networks. To address these gaps, we propose to develop a framework (called RAVEN) for extracting, assessing, and forecasting evolving narratives in social networks. Our approach seeks to 1) identify and characterize the key narrative frames and structures, grounded in group interests, ideology, and morals, that shape perception and response to concrete focal events, 2) detect related focal events and quantify their impact on narratives, and 3) analyze how narratives interact within dynamic social networks. To tackle these distinct but essentially inter-related challenges, we exploit a variety of new and established AI and NLP techniques. To identify and characterize key narrative frameworks, we leverage deep language models, named entity recognition, topic modeling, and stance and emotion analysis. To detect focal events, we employ advances in event extraction and relation extraction. We will establish the feasibility of RAVEN in Phase I by demonstrating an initial implementation of an analytics pipeline that identifies, characterizes, and tracks evolving narratives along with the focal events that drive narrative change. To this end, we will pursue three main objectives. Objective 1 aims to build upon the foundational narrative extraction methodology developed by our Co-Investigator, Dr. Nitin Agarwal, augmenting narrative representations using state-of-the-art emotion detection and stance detection algorithms. This will allow us to prioritize narratives that express social, moral, or ideological responses and commitments. Objective 2 seeks to extract focal events and link them to their associated narratives via shared actors, entities, and topics. Objective 3 seeks to model the spread and evolution of narratives across online communities (networks) and sub-networks, identifying the specific network properties that predict narrative propagation.