Community detection and Higher-order Link Prediction

Abstract Community detection algorithms utilize graphs as input data, representing networks that are often incomplete. Many networks, whether obtained through collection or inference, suffer from missing edges, either inherent to the dataset or due to sampling limitations. Networks evolve over time, and it is imperative to predict future links. An excellent example is the spread of epidemics, where such predictive knowledge could potentially save lives. The standard approach involves examining the dyadic nature of links, as it is the perspective offered by graph frameworks, and attempting to anticipate future edges. However, being part of a large system entails multiple interactions that the link paradigm does not clearly elucidate. The idea presented in this paper is to employ higher-order link prediction to achieve robust results regarding the community detection problem. Finally, we evaluate our method on a series of ground truth networks and artificial networks. Examples from the LFR framework demonstrate that our method improves the community detection problem on incomplete networks.