Quantifying the impact of scientific collaboration and papers via motif-based heterogeneous networks

Structured measurements have been widely used to measure the impact of scholarly entities based on scholarly networks. Existing methods use heterogeneous scholarly networks and the PageR-ank algorithm to quantify the impact of scientific collaboration. However, by ignoring important higher-order relationships in citation networks, the impact of scientific collaboration is quanti-fied by relying on first-order relationships, which leads to indistinguishable self-citations. In this paper, to address these shortcomings, we propose a Motif-based Scientific Collaboration Impact Rank framework, named as MSCIRank, which leverages the triangular motifs from the recon-structed collaboration-citation networks and integrates the first-order and higher-order relation-ships in the PageRank algorithm to quantify the impact of scientific collaboration and scholarly papers. MSCIRank consists of two models, i,e, linear and non-linear. Extensive experiments have demonstrated the effectiveness of MSCIRank. The experimental results show that MSCIRank is better than SCIRank in identifying Nobel Prize papers in terms of Recall. The MSCIRank model can weaken or strengthen the impact of self-citation. Linear MSCIRank is consistent with Pareto's principle, while non-linear MSCIRank is inconsistent. In addition, the average impact of pairs of co-authors with high impact in the linear MSCIRank is much higher than that in the non-linear MSCIRank.