Abstract:

[Purpose/Significance] In the current landscape of global scientific competition, understanding how top researchers reuse existing knowledge is critical for talent development and research policy. Prior studies has mainly examined high-performing scholars via macro-level indicators such as age, education, and institution, but has not penetrated the micro-level behavioral processes of knowledge reuse in actual practice. To fill this gap, this study adopts a lexical semantic function perspective that distinguishes knowledge into four functional types - method, problem, background, and object - based on their specific semantic roles within academic texts. This framework enables us to assess how scholars reuse these types and the independent and joint effects of each type on publication output. The findings provide empirical evidence that can guide targeted training, research management, and knowledge resource allocation, thereby contributing to both theory on knowledge management and practical efforts to foster scientific innovation. [Method/Process] This study collected 15 471 articles (2020-2024) from 18 CSSCI-indexed library and information science (LIS) journals in China, extracting author, keyword, title, and abstract fields. To classify keywords into functional types, this study implemented a hybrid pipeline that combined template-based labeling, large language model (DeepSeek-R1) validation, and Bert‑BiLSTM classification. After annotation, this study applied K‑means clustering to aggregate knowledge items within each functional category into thematic clusters based on co-occurrence patterns with other functional types, thereby transforming individual keywords into broader knowledge units that better capture scholars' thematic continuity. For 85 prolific scholars identified by Price's Law, this study quantified reuse intensity using the Herfindahl-Hirschman index and derived a knowledge focus metric (ranging from 0 to 1, with higher values indicating greater concentration) for each of the four functional categories. This study then performed Spearman correlation and multiple linear regression analyses, including interaction terms, to examine the relationships between knowledge focus and publication frequency, while controlling for multicollinearity and testing synergistic effects among knowledge types. [Results/Conclusions] Knowledge reuse is nearly universal among prolific LIS scholars: 96.5% exhibit high reuse (knowledge focus≥0.5) for background knowledge, whereas only about 30% show similar levels for method, problem, or object knowledge. This dominance indicates that sustained thematic engagement serves as the primary foundation for high productivity. Most scholars (49.4%) reuse knowledge across exactly two functional categories, revealing a "dual‑dimension anchoring" strategy that combines stable background with one other dimension. Correlation analyses show that knowledge focus in all four categories is significantly positively associated with publication frequency, with background knowledge having the strongest correlation (r=0.657, p=0.005), followed by object (r=0.325). Regression confirms significant independent positive net effects for each type. More importantly, interaction analysis reveals a strong synergistic effect between method and background reuse (coefficient=94.460, p=0.001), indicating that scholars who maintain high methodological consistency within a coherent thematic domain achieve substantially greater output than the sum of individual effects. No other interactions are significant. These findings suggest that background knowledge accumulation should be prioritized in early-career training, and that methodological instruction should be integrated with domain applications to exploit the synergy. Limitations of this study include neglecting knowledge half‑life and potential contextual variability in functional classification. Future work should incorporate temporal dynamics and refine recognition using large language models for more adaptive cross‑disciplinary analysis.

Key words: knowledge reuse behavior, prolific scholars, Herfindahl index, lexical function recognition, literature knowledge clustering