interdisciplinary collaboration

Quantifying and leveraging interdisciplinary synergies in higher STEM education remains challenging. Method: This study develops a quantitative evaluation framework grounded in information theory and educational theory, leveraging large-scale official Korean educational data. Course syllabi are standardized and mapped onto discipline-specific knowledge graphs using large language models. A novel metric—the “Curricular Synergy Score”—is introduced to quantify interdisciplinary alignment, informing the design of both single-discipline and dual-discipline integrated STEM curricula. Results: Engineering disciplines emerge as central hubs of interdisciplinary synergy, with foundational natural sciences forming a supportive periphery; intra- and inter-engineering combinations exhibit significantly higher synergy than other pairings. The study yields a scalable, empirically validated curriculum paradigm—applicable to single-domain or dual-domain STEM programs—that provides both methodological rigor and actionable evidence for interdisciplinary curriculum design.

Cross-disciplinary cold-start knowledge tracing (CDCKT) faces severe challenges due to extremely sparse interaction data in the target discipline and the absence of overlapping entities, rendering existing mapping-based approaches inadequate for modeling complex cross-disciplinary knowledge associations. To address this, we propose AdaptKT—a novel adaptive knowledge transfer framework integrating Mixture of Experts (MoE) and Generative Adversarial Networks (GAN). Its core innovation lies in leveraging source-discipline knowledge-state clusters as learnable cross-domain semantic bridges, enabling fine-grained feature alignment and disentanglement without shared entities. AdaptKT synergistically combines pretrained representations, clustering-guided MoE gating, and an adversarial discriminative module to significantly enhance knowledge-state modeling under few-shot target-discipline settings. Evaluated across 20 extreme cross-disciplinary cold-start scenarios, AdaptKT consistently outperforms state-of-the-art methods, demonstrating strong generalizability and practical applicability.

To address the high construction cost, uneven interdisciplinary coverage, and delayed updates of research-domain ontologies, this paper proposes the first large language model (LLM)-based framework for automated, multi-disciplinary ontology generation. We introduce PEM-Rel-8K—a high-quality, manually curated relation extraction dataset spanning biomedical, physics, and engineering domains—and systematically evaluate LLMs under zero-shot, chain-of-thought prompting, and fine-tuning paradigms for cross-domain semantic relation identification. Experimental results demonstrate that models fine-tuned on PEM-Rel-8K achieve state-of-the-art performance across all three disciplines, significantly outperforming existing baselines while exhibiting robust cross-domain transferability. This work establishes a scalable, low-cost paradigm for the automated construction and dynamic evolution of scientific knowledge graphs.

Existing approaches struggle to assess, at a fine-grained level, how citations in interdisciplinary research substantively integrate ideas from multiple fields. This work proposes a citation-purpose classification framework tailored for interdisciplinary scholarship, combining manual annotation, citation context analysis, and qualitative categorization to construct the first quantifiable system that measures both the depth and significance of citation engagement. Validated on publications at the intersection of natural language processing and computational social science, the framework not only uncovers actual patterns of interdisciplinary citation usage but also provides an actionable metric for evaluating citation quality.

This study identifies a structural imbalance in interdisciplinary data sharing: high reuse rates in STEM fields contrast sharply with low adoption in humanities and social sciences, while persistent undercitation of datasets impedes evidence-based policy and infrastructure development. Leveraging full-text PubMed articles, we construct the first multidisciplinary dataset—enabling simultaneous identification of data mentions and classification of data-related intents—by integrating natural language processing, full-text pattern recognition, cross-disciplinary bibliometrics, and time-series modeling. Key findings include: (1) a marked acceleration in data publication post-2012; (2) highest data publishing activity in business/management and creative arts, yet highest reuse in biological and agricultural sciences; and (3) consistently low dataset citation rates, revealing critical bottlenecks in discoverability and format interoperability. These empirically grounded insights advance data governance frameworks and support formal recognition of datasets as independent scholarly outputs.

This study investigates how the organizational structures of Library and Information Science (LIS) schools in the United States shape the evolution of their research agendas. Drawing on a corpus of 14,705 publications by 1,264 faculty members across 44 institutions, and employing computational methods such as word embeddings and topic modeling, the research systematically identifies three major research dimensions for the first time. It reveals that the LIS field exhibits a diversified structure dominated by “Human-Centered Technology” (HCT), challenging the prevailing narrative of computational dominance. The findings indicate that 51.4% of LIS schools are shifting toward HCT; departments affiliated with computer science tend to concentrate on computationally intensive research, whereas standalone information schools demonstrate the highest research diversity. These insights offer empirical grounding for strategic planning and policy-making in the discipline.

This study addresses the fragmented and unevenly advertised landscape of academic policy engagement opportunities, which hinders researchers’ effective participation in public policymaking. It presents the first systematic integration of multinational policy engagement data into a structured database and introduces an intelligent recommendation framework that aligns scholars’ publication records with opportunity descriptions through natural language processing and semantic matching algorithms. The analysis reveals significant disparities in opportunity distribution across countries and disciplines, demonstrates a positive correlation between institutional publication volume and high-confidence matching rates, and shows that domain-specific research expertise can offset disadvantages associated with lower output volume. The proposed framework enables comparative analyses across nations, fields, and institutions, offering a scalable infrastructure to bridge the gap between scientific research and policy engagement.

This study investigates how domain-specific metadata schemas can be effectively integrated with the generic DataCite schema to enhance metadata quality and interoperability in research data repositories. Through structural comparisons, cross-schema mapping analyses, and workflow evaluations of metadata records from eight repositories in the earth and social sciences, the research reveals how disciplinary characteristics influence the completeness of DataCite records. Findings indicate that discrepancies between schemas stem primarily from differing modeling philosophies rather than expressive capacity. While optimized cross-schema mappings significantly improve metadata quality, the diversity of repository workflows also critically affects record completeness. Building on these insights, the study proposes a strategy that leverages the complementary strengths of domain-specific and generic schemas, offering practical guidance for fostering interdisciplinary data sharing.

Bridging the gap between general biomedical knowledge and actionable, testable hypotheses for specific experimental or clinical contexts remains a critical challenge. This work proposes SCENE, a novel framework that formalizes knowledge contextualization as an iterative search process through a dual-layer multi-agent architecture to deeply integrate knowledge-driven and data-driven reasoning. The upper-layer agent generates search directions and anchors relevant data patterns, while the lower-layer agent leverages knowledge graph guidance and multi-objective optimization to produce verifiable propositions that balance evidential strength with empirical support. Evaluated in real-world settings, SCENE successfully identified patient subgroups with heterogeneous treatment effects in clinical trials and discovered perturbation contexts with high target-response alignment in the LINCS L1000 study, significantly outperforming existing baselines. The generated hypotheses exhibit strong traceability, reproducibility, and expert verifiability.

This study addresses key challenges in applying large language models (LLMs) to the social sciences and humanities (SSH), including disciplinary heterogeneity, limited access to multilingual scholarly literature, and insufficient evaluability of outputs. To overcome these issues, the work proposes a domain-adaptive framework that uniquely integrates knowledge graphs with multilingual academic corpora, deeply coupling domain sensitivity, regulatory compliance, and generative AI for the first time. The approach leverages knowledge graph embeddings, retrieval-augmented generation, multilingual fine-tuning, and ethical compliance mechanisms, all aligned with the LLMs4EU evaluation protocol, to develop a trustworthy, traceable, and responsible SSH-specific model. Experimental results demonstrate strong performance across retrieval, summarization, traceability, and hallucination detection metrics, with qualitative validation by digital humanities experts confirming its scholarly applicability and reliability.