Uncovering Diffusion in Academic Publications Using Model-Driven and Model-Free Approaches

Information spreads across heterogeneous social systems, and the underlying network structures are hard to collect or define. The goal of this paper is to estimate macro-level information diffusion using time-series activity sequences of heterogeneous populations without the need to know detailed network structures. We propose a consistent way of understanding dynamic influence among populations with both model-driven and model-free approaches. As a real-word example, we focus on computer science publications for uncovering research topic diffusion patterns across different sub domains. As a result, estimated diffusion patterns, obtained from the two approaches, exhibit similar information pathways but with different perspectives on diffusion, which in conjunction can help to obtain a more coherent overall picture of diffusion dynamics than either approach alone. We expect that our proposed approaches can help quantify and understand macro-level diffusion across target regions in various real-world scenarios and provide ways of inferring diffusion patterns from time-series real data.