Introducing LENS: A Methodology for the Layered Exploration of Narrative Structures in Opaque Social Platforms

Structurally opaque platforms are digital environments where users' interaction graphs and information diffusion traces are partial, unreliable, or unavailable. This paper introduces LENS (Layered Exploration of Narrative Structures), a multiscale framework for inferring narrative structure under such conditions. LENS defines a reproducible and transparent methodological protocol linking corpus construction and relational inference through expert-guided signal identification, cross-platform pivoting, and semantic enrichment with embeddings, named entities, and sentiment. The resulting semantic space is used to detect temporal anomalies, typify communities, infer directional influence from recurrent semantic anticipation, and model adversarial associations. Its recursive Zoom-In procedure isolates anomalous windows and re-clusters them at higher resolution, separating overlapping events and removing exogenous noise. We validate LENS on 1,216,335 messages from 53 Spanish-language Telegram channels collected between 24 February 2020 and 7 September 2025. Results distinguish discursive silos from a public square; within the latter, the inferred influence graph reveals an asymmetric lead--lag hierarchy in which broadcaster channels systematically precede amplifiers, while the main sink node absorbs nearly 35 times more semantic volume than it originates. In two validation cases, Zoom-In achieves robust separation with a Silhouette index above 0.6, supporting detection of weak signals.