<p>Knowledge diversity plays a pivotal role in shaping research outputs in scientific collaboration. However, most previous studies measure knowledge diversity at the article level and implicitly treat collaboration members as a homogeneous whole, which may lead to aggregation bias. This study proposes a dual-perspective framework for measuring knowledge diversity in scientific collaboration, distinguishing between the individual perspective—each collaborator is treated as a separate individual—and the collective perspective—all collaborators are treated as an integrated entity. Using articles published in <i>Nature</i> between 2000 and 2020 as the empirical dataset, knowledge diversity is measured across three dimensions: variety, balance and disparity, and its associations with citation impact and disruption are examined. The results reveal divergent patterns across the individual and collective perspectives: individual knowledge variety is positively associated with both impact and disruption, whereas collective knowledge variety is negatively associated with both citation impact and disruption. Knowledge balance shows no stable association with impact, but its relationship with disruption differs across perspectives. Collective knowledge disparity is positively associated with both impact and disruption, whereas individual knowledge disparity is positively associated with impact but negatively associated with disruption. These findings highlight the importance of distinguishing between the two perspectives of knowledge diversity and provide a refined lens for optimizing team composition and knowledge integration in scientific collaboration.</p>