Temporal Narrative Monitoring in Dynamic Information Environments

David Farr, Stephen Prochaska, Jack Moody, Lynnette Hui Xian Ng, Iain Cruickshank, Kate Starbird, Jevin West · 2026

Read Temporal Narrative Monitoring in Dynamic Information Environments by David Farr, Stephen Prochaska, Jack Moody, Lynnette Hui Xian Ng, Iain Cruickshank, Kate Starbird, Jevin West (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-17617.txt.

Introduction

Farr and co-authors build a system that tracks how crisis narratives shift over time. Snapshot classifiers miss the way stories in a fast-moving information space evolve. Their framework embeds posts, clusters them by density, and links clusters across rolling time steps. Narratives then persist, adapt, split, or fade inside one shared space. A real crisis case showed tight clusters and mixed life spans, they report. The tool aids monitoring and decisions.

Catalog record

CREATORDavid Farr, Stephen Prochaska, Jack Moody, Lynnette Hui Xian Ng, Iain Cruickshank, Kate Starbird, Jevin West
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26955
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.17617

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Temporal Narrative Monitoring in Dynamic Information Environments David Farr, Stephen Prochaska, Jack Moody, Lynnette Hui Xian Ng, Iain Cruickshank, Kate Starbird, Jevin West Abstract Comprehending the information environment (IE) during crisis events is challenging due to the rapid change and abstract nature of the domain. Many approaches focus on snapshots via classification methods or network approaches to describe the IE in crisis, ignoring the temporal nature of how information changed over time. This work presents a system-oriented framework for modeling emerging narratives as temporally evolving semantic structures without requiring prior label specification. By integrating semantic embeddings, density-based clustering, and rolling temporal linkage, the framework represents narratives as persistent yet adaptive entities within a shared semantic space. We apply the methodology to a real-world crisis event and evaluate system behavior through stratified cluster validation and temporal lifecycle analysis. Results demonstrate high cluster coherence and reveal heterogeneous narrative lifecycles characterized by both transient fragments and stable narrative anchors. We ground our approach in situational awareness theory, supporting perception and comprehension of the IE by transforming unstructured social media streams into interpretable, temporally structured representations. The resulting system provides a methodology for monitoring and decision support in dynamic information environments. Fig. 1 : System architecture for temporal narrative monitoring. External events drive ingestion into a shared semantic space where temporal clustering produces evolving narrative clusters in a semantic state space. Centroid-based monitoring and generative summaries support analyst review and adaptive feedback. I Introduction Understanding crisis information environments is challenging because online discourse changes rapidly as narratives form, compete, and dissipate in response to unfolding events. Existing analytical approaches commonly rely on classification-based NLP pipelines or network-based analyses of user interactions 13 ; 1 . While effective in relatively stable settings, these methods struggle in dynamic environments where narratives evolve quickly, producing continual distributional change in both content and discourse patterns 8 ; 11 . Classification workflows typically require predefined labels, implicitly assuming stable categories and prior knowledge of relevant narratives 13 . Network-based approaches instead emphasize interaction structure, providing insight into information flow but limited visibility into how narrative meaning itself evolves over time 12 . Together, these limitations motivate treating narratives not as static labels or latent topics, but as evolving system-level entities. In this work, we present a system-oriented framework that models narratives as temporally evolving semantic structures without requiring prior label specification. We make three contributions: (1) a scalable NLP system for detecting and tracking narratives in non-stationary social media streams, (2) a clustering-based temporal modeling approach enabling measurement of narrative emergence and drift, and (3) an empirical evaluation demonstrating system behavior during a real-world crisis event in Venezuela. Unlike prior topic modeling or clustering approaches that produce independent temporal snapshots, the proposed framework treats narrative structure as a persistent system state evolving within a shared semantic space. This enables measurement of narrative continuity, drift, and stability as observable system properties, transforming discourse monitoring from static analysis into dynamic system monitoring. II Background and Related Work This work integrates concepts from multiple research areas, including situational awareness, natural language processing, and computational social science. We draw on these traditions to design a system that supports the representation and interpretation of complex, evolving information environments. II-A Situational Awareness Our work is grounded in the situational awareness literature, particularly the three-level model of situational awareness proposed by Endsley 6 . In this framework, Level 1 situational awareness involves the perception of elements in the environment, including their status, attributes, and dynamics. Level 2 situational awareness concerns the comprehension of these elements and their significance within a broader context. Level 3 situational awareness involves the projection of future states based on an understanding of current conditions and trends 6 . Endsley defines situational awareness as “the perception of the elements in the environment within a volume of time and space, the comprehension of their meaning, and the projection of their status in the near future” 6 . This model has been widely used to reason about how information processing supports decision making in complex, time-sensitive environments. We position our system as supporting the first two levels of situational awareness by improving perception and comprehension of the information environment. By structuring large-scale, dynamic social media data into temporally evolving narrative representations, the system provides a mechanism for filtering, organizing, and summarizing salient information. This supports information processing by reducing cognitive load and enabling analysts to observe how narratives emerge, persist, and change over time, rather than requiring direct engagement with raw data streams. Our methodology is designed to support perception and comprehension through high-volume information pre-processing and prioritization via temporal windows and persistent cluster identity. Generative summarization is used as an abstraction mechanism to support interpretation of salient discourse, helping mitigate disruptions to information pre-processing, prioritization, confidence assessment, and interpretation that are known to degrade situational awareness and decision performance 5 . Our approach also avoids the trap of supervised-based systems in having to pre-define topics, by instead surfacing emergent clusters in a way that enables human sense-making. Our approach addresses these challenges by providing a structured, temporally grounded representation of discourse without assuming predefined categories or static interpretations. II-B Natural Language Tools Applied to Social Media Data A wide range of natural language processing techniques have been applied to social data to support large-scale analysis and monitoring. These approaches generally fall into three categories: classification-based methods, network-based analyses augmented with textual features, and temporal approaches designed to capture change over time. While each category offers useful capabilities, they also exhibit limitations when applied to dynamic information environments where narratives emerge, evolve, and dissipate among large volumes of data. II-B 1 Classification Numerous approaches apply large language models and fine-tuned neural models to classify social media data, including tasks such as stance detection, misinformation identification, and topic classification 7 ; 13 . These methods span supervised and weakly supervised pipelines as well as zero-shot and few-shot large language models 3 . When label spaces are well defined and stable over time, classification-based approaches can efficiently characterize large volumes of data. However, such approaches typically rely on predefined labels or coarse-grained categories (e.g., for , against , neutral ), which can obscure important nuance in complex information environments 10 . In rapidly evolving settings, narratives often emerge and transform in ways that are difficult to anticipate a priori, making static label sets challenging to define and maintain 8 . Moreover, classification pipelines generally operate on individual data points or snapshots in time, providing limited visibility into the changes in discourse over time or narrative lifecycles. II-B 2 Network-Based Methods Network-based approaches to social media analysis model information environments through the relational structure of user interactions, commonly using retweet, reply, or co-engagement graphs to identify communities, map information flows, and detect coordinated behavior 12 ; 1 . These methods are particularly well-suited to identifying structural properties of discourse, such as community polarization, influence hierarchies, and cross-platform diffusion 11 . By representing the information environment as a graph in which nodes are actors and edges are interactions, network-based analyses can reveal who amplifies what content and how narratives propagate through connected populations. Despite these strengths, network-based approaches face inherent limitations in dynamic monitoring settings. Because interaction structure

Read the complete text