Abstract
While information theory is widely used to quantify knowledge combinations, the fundamental principles guiding reference selection in science remain largely unexplored. This study analyzes a large-scale journal citation network to introduce and empirically validate a principle we term "Mediating Similarity". We posit that a journal's reference list acts as a strategic cognitive bridge, creating a more efficient informational path from its specific research identity to the broader scientific landscape. Using information-theoretic measures and computational experiments, we tested this principle and its underlying mechanisms. Our findings provide robust, multi-level evidence. First, we confirm the universality of the principle, showing that the mediated path through references is consistently more efficient than the direct path for thousands of journals. Second, perturbation experiments reveal a dual mechanism guiding reference selection: real-world reference portfolios are not merely collections of relevant works, but are synergistically optimal combinations that vastly outperform randomly assembled alternatives. This global optimization, however, operates as a robust "satisficing" strategy, balancing the search for an ideal cognitive path with the practical constraints of scientific discovery. Collectively, these findings reframe reference behavior as a strategic process of navigating a cognitive energy landscape, where journals selectively curate references to enhance their integrative capacity and innovative potential.