Intelligent Construction and Application of Citizen-Science-Data-Driven Knowledge Graphs for Popular Science Publishing: The Case of the Ecological Domain

Citizen science data, generated through large-scale public participation in scientific observation, holds unique potential for popular science publishing in both scale and engagement. Yet its heterogeneity in scientific credibility and knowledge organization renders direct transformation into qualified content problematic. This paper focuses on citizen science data in the ecological domain, systematically examining its applicability and proposing an integration framework. The paper first establishes a principled two-tier classification: "verified scientific data," sourced from structured projects with standardized protocols and expert validation, and "individual narratives pending verification," encompassing personal observations, emotional experiences, and aesthetic reflections. The former can serve as core sources of scientific facts; the latter should only enter publishing as contextual material and empathy triggers—not as scientific knowledge entities. This stratified approach, essential for maintaining scientific integrity, constitutes the epistemological foundation for all subsequent technical design. Building on this classification, the paper systematically compares research-oriented and popular science publishing knowledge graphs, revealing essential differences across objectives, target users, knowledge granularity, verification mechanisms, and product forms. A central insight emerges: a popular science publishing knowledge graph is not a simplified version of its research counterpart, but an entirely new artifact with its own design philosophy and product logic. Accordingly, a five-layer architecture is proposed—comprising data source, knowledge extraction, knowledge fusion, schema, and storage/service layers—to be led by publishing institutions through multi-party collaboration. At the technical core lie two carefully designed mechanisms. The first is a multi-dimensional credibility scoring model evaluating each knowledge unit along four weighted dimensions: source authority, cross-validation strength, spatiotemporal consistency, and user historical reliability, producing a quantifiable score that drives differentiated review workflows. The second is an "event-mediated mapping" mechanism for narrative association, whereby individual narratives are abstracted into event nodes carrying temporal, spatial, and emotional attributes before being precisely linked to verified scientific entities. This design maintains an arm's-length relationship between narrative material and scientific fact, while a front-end source visualization further safeguards against conflating verified knowledge with personal experience. The paper further elaborates four core capabilities—relational organization, evolutionary reasoning, hierarchical representation, and narrative expression—and devises four corresponding intelligent publishing models: dynamic iteration, precision matching, immersive narration, and cross-media derivation. Three specific risks are identified—narrative impoverishment, spurious correlation inference, and audience cognitive fragmentation—with targeted countermeasures embedded within a dual technical-institutional safeguard mechanism. As a prospective theoretical exploration, the proposed technical pathways remain conceptual, their feasibility and viability awaiting empirical validation through prototype development and industry collaboration. Future work will extend the framework to other citizen science-intensive domains such as astronomy and climate science. The framework's core value lies in providing a systematic, discussable, and revisable reference for the data-driven transformation of popular science publishing.