Temporal Motifs in Heterogeneous Information Networks

Network motifs are crucial building blocks of understanding and modeling complex networks for their capacity in characterizing higher-order interactions. Meanwhile, heterogeneous information networks (HINs) are ubiquitous in real-world applications, which o‰en come with rich temporal information. We are hence motivated to study temporal motifs in the context of heterogeneous information networks. With examples from real-world datasets, we demonstrate HIN motifs can be armed with substantially more discriminability by incorporating temporal information. Furthermore, counting temporal HIN motif instances in large-scale networks is time consuming. We therefore develop ecient counting algorithm for the HIN motifs that are of the most interests in the literature. Empirical observations in the experiment have shown that interesting motif instances can be identi€ed from large-scale HINs thanks to the improved discriminability of temporal HIN motifs, and the proposed ecient counting algorithm enjoys linear complexity that is multiple orders of magnitude faster than the baseline method in three real-world HINs.