Thematic PageRank: A Hybrid Approach for Evaluating Node Importance and Discipline Similarity in Interdisciplinary Networks

As scientific research becomes increasingly interdisciplinary, it is essential to quantify both the structural importance of disciplines and the topical proximity among them in cross-disciplinary networks. However, conventional PageRank mainly relies on link structure and therefore may overlook disciplines that are weakly connected structurally but closely related in terms of research themes. To address this limitation, we construct an interdisciplinary network dataset centered on computer science using metadata from OpenAlex, where nodes represent disciplines and edges represent co-occurrence relationships among disciplines in scholarly records. We further extract discipline-level topic representations from paper abstracts and keywords and propose a hybrid ranking method, namely Thematic PageRank (TPR), which integrates structural connectivity with topic similarity to evaluate node importance more comprehensively. In addition, we define multiple similarity measures to examine structural, thematic, and hybrid associations among disciplines. To validate the proposed method, we compare TPR with several baseline ranking algorithms, including standard PageRank, Weighted PageRank, Topic-Sensitive PageRank, HITS, LeaderRank, and Node2Vec. We also design ablation experiments to test the contribution of the structural branch, topic branch, and fusion mechanism. Experimental results show that TPR is more effective in identifying interdisciplinary disciplines at the boundary of the network, while the hybrid similarity measure provides stronger clustering performance than single-dimension measures. The proposed framework offers a more systematic and interpretable approach for evaluating disciplinary importance and similarity in interdisciplinary networks.