Scholarlink

<p>Scientific research and development are significant drivers of innovative organizations. The selection of high-quality research topics to invest in directly contributes to the prospects of R&amp;D-based organizations. The emergence and evolution of the research topic is a complex process involving nonlinear interactions among several factors. Machine learning approaches are adept at fitting high-dimensional complex interactions, and we selected the eight classifiers to identify high-quality research topics. By training these models using high-quality research topic datasets, we found that the Gradient Boosting classifier exhibits superior performance in detecting high-quality research topics, as evidenced by its F1-score of 94.45%. Further, we conducted SHapley Additive exPlanations (SHAP) analysis to interpret the model’s decision-making process. Our findings demonstrate that high-quality research topics exhibit three distinctive signatures: (1) rapid engagement by the scientific community, evidenced by a sharp rise in publications and citation rates within 2–3 years of emergence; (2) disproportionate early adoption by highly cited authors; and (3) a characteristic network topology featuring a sparse, ego-centered structure with limited direct connections to existing topics and weak interlinkages among neighboring domains.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Identifying high-quality research topics in R&D organizations using explainable machine learning

  • 期刊论文

  • 发表日期：

    2025年11月27日

查看全文
  • Jinqing Yang,
  • Jiming Hu,
  • Jiaqi Li

摘要

Scientific research and development are significant drivers of innovative organizations. The selection of high-quality research topics to invest in directly contributes to the prospects of R&D-based organizations. The emergence and evolution of the research topic is a complex process involving nonlinear interactions among several factors. Machine learning approaches are adept at fitting high-dimensional complex interactions, and we selected the eight classifiers to identify high-quality research topics. By training these models using high-quality research topic datasets, we found that the Gradient Boosting classifier exhibits superior performance in detecting high-quality research topics, as evidenced by its F1-score of 94.45%. Further, we conducted SHapley Additive exPlanations (SHAP) analysis to interpret the model’s decision-making process. Our findings demonstrate that high-quality research topics exhibit three distinctive signatures: (1) rapid engagement by the scientific community, evidenced by a sharp rise in publications and citation rates within 2–3 years of emergence; (2) disproportionate early adoption by highly cited authors; and (3) a characteristic network topology featuring a sparse, ego-centered structure with limited direct connections to existing topics and weak interlinkages among neighboring domains.