研究前沿视角下的潜在科研创新团队识别研究OACHSSCD

[Purpose/Significance]As scientific research becomes increasingly collaborative,interdisciplinary,and knowledge-intensive,early identification of research teams with strong innovation potential has become important for research management and resource allocation.Existing team identification studies have mainly relied on co-authorship net-works and other structural indicators to detect established collaboration groups.However,such approaches often overlook the semantic content of research outputs and therefore have limited ability to identify emerging groups associated with rapidly evolving research fronts.To address this limitation,the paper proposes a framework for identifying potential research innovation teams from the perspective of research fronts.[Method/Process]Taking the field of high-efficiency perovskite solar cells as an example,the paper developed a framework integrating innovation evaluation and community detection.Full-text data were collected from two sources:citing papers of core papers in a research front and subsequent publications by the authors of those core papers.After data cleaning and parsing,22,738 papers containing titles,abstracts,and con-clusions were retained.The GLM-4.5 large language model was then used to extract problem and method knowledge ele-ments from scientific texts.Based on the novelty of problem elements,method elements,and their combinations,the paper evaluated paper-level innovation and selected 18,686 highly innovative papers.For the authors of these papers,con-tent similarity was measured using TF-IDF and Word2Vec based on the extracted knowledge elements,while network simi-larity was calculated using the Common Neighbors index with Jaccard normalization.The two similarity matrices were inte-grated through weighted fusion,and the Louvain algorithm was applied to the weighted author-author matrix to identify potential research innovation teams.Its performance was also compared with that of the GN and spectral clustering algo-rithms.[Result/Conclusion]The empirical analysis identified 535 potential research innovation teams in the selected field.The results show that the proposed method can not only effectively capture the core scientific research forces in the field,but also uncover potential collaborative groups across institutions and regions.The integration of semantic similarity and network structural proximity improves the identification of potential collaborative relationships,and the Louvain algorithm performs better than the comparison algorithms in both modularity and computational efficiency.These findings suggest that a research-front-oriented approach provides an effective means of identifying potential scientific innovation teams by jointly considering innovation quality,knowledge relatedness,and collaboration possibility.The limitations of the paper lie in its focus on a single field and its relatively static treatment of team formation.Future research would extend the frame-work to additional disciplines and incorporate dynamic evolution analysis.