A network-based known-unknown model for predicting scientific impact from topic selection strategies

Abstract The selection of research topics is a fundamental driver of scientific progress, shaping both individual careers and the collective evolution of knowledge. This study introduces the Known–Unknown Model (KUM), a parsimonious single-parameter framework designed to explain how scientists select and transition between research topics over the course of their careers. The KUM model conceptualizes topic choice as a binary decision: either to exploit familiar domains (“known”) or to explore unfamiliar ones (“unknown”), thereby formalizing the enduring tension between depth and breadth that characterizes scientific inquiry. Using large-scale publication data from the American Physical Society (APS), we show that KUM successfully reproduces key empirical regularities in individual topic repertoires, including patterns of specialization, diversification, and topic persistence. Moreover, KUM achieves higher predictive accuracy for long-term academic impact—measured by both the H-index and the Q-index—than several established benchmark models. The results indicate that balanced exploration–exploitation strategies are systematically associated with greater sustained influence, whereas extreme conservatism or excessive novelty-seeking tends to be suboptimal. Beyond its empirical performance, the model offers a transparent and interpretable framework for understanding research behavior, with implications for science policy, funding strategies, and career planning, particularly for early-career researchers navigating uncertainty and risk in topic selection.