An Attract-Repel Decomposition of Undirected Networks
Abstract: Dot product latent space models are a standard method in many areas ranging from social network analysis to computational biology. Such models have issues modeling graphs which include unclosed triangles such as social networks which include latent heterophily (i.e. cases where opposites attract) or co-occurrence graphs which have substitutes (items which occur in similar contexts but not together). We show a minimal expansion to the dot product model which includes both homophily (attract) and heterophily (repel) latent forces. Beyond simply fitting the data, we discuss how to use the AR spaces produced to more deeply understand real networks allowing analysts to measure the latent heterophily in social network formation, detect substitutes in co-occurrence networks, or perform exploratory analysis for candidates for inhibition / activation relationships in systems biology.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Jaakko_Peltonen1
Submission Number: 315
Loading