Reflections and Path Exploration on Building Industry Datasets for Sci-Tech Publishers Driven by "Model-Data Resonance"

Competition in the artificial intelligence industry is increasingly shifting from a contest of general-purpose capabilities to deep industry applications, yet the insufficient supply of high-quality industry datasets has become the key bottleneck constraining the "AI+" empowerment of diverse industries. Science and technology publishers (STPs), with their long-accumulated professional knowledge resources and deep-rooted industry networks, are well positioned to become core suppliers, yet they simultaneously face multiple challenges including lagging strategic awareness, structural supply-demand mismatches, high processing costs and technical difficulties, the absence of shared tools and standards, underdeveloped circulation mechanisms, and a critical shortage of interdisciplinary talent. This paper systematically reviews recent national policies on "AI+" and grounded in the core concept of "model-data resonance", elucidates its positive feedback loop: high-quality industry data train more models with stronger domain knowledge, and the application of these models in turn generates new scenarios and produces fresh data, which then feeds back to further optimize the models. The paper clarifies the definition of industry datasets, which encompass multimodal and multidimensional data forms across three hierarchical layers—basic data modalities, specialized professional data forms, and knowledge organization structures—and analyzes the driving factors, role positioning, distinctive advantages, and practical challenges that STPs face in constructing such datasets. Building on this analysis, the paper proposes nine concrete pathways at both strategic and implementation levels. Strategically, these include raising strategic awareness of the data economy, strengthening policy research to align with national directives, clearly defining priority industry directions, and fostering a deeply integrated ecosystem that brings together government, industry, academia, research, and application stakeholders. On the execution front, the pathways consist of maintaining the core traditional publishing business while leveraging it as a foundation, adopting a "three-step method" for data identification, annotation, and iterative validation to optimize the development process, co-developing shared tools and standards in collaboration with industry partners, assembling interdisciplinary professional teams that combine editorial, technical, and domain expertise, and placing special emphasis on constructing industry ontology datasets to ensure semantic interoperability and long-term reusability. The paper contends that 2026 marks a pivotal year for China's AI trajectory—the transition from general-purpose conversational AI to practical industry applications. For science and technology publishers, this represents a strategic window of opportunity. Guided by the principle of "model-data resonance", these publishers should accelerate their transformation from traditional content producers to data service providers, converting their accumulated intellectual capital from professional publishing into a distinctive competitive advantage in the intelligent age, thereby providing robust support for building China's independent knowledge system and enhancing the nation's overall scientific and technological competitiveness.