Graph Summarization Methods and Applications

Abstract

While advances in computing resources have made processing enormous amounts of data possible, human ability to identify patterns in such data has not scaled accordingly. Efficient computational methods for condensing and simplifying data are thus becoming vital for extracting actionable insights. In particular, while data summarization techniques have been studied extensively, only recently has summarizing interconnected data, or graphs, become popular. This survey is a structured, comprehensive overview of the state-of-the-art methods for summarizing graph data. We first broach the motivation behind, and the challenges of, graph summarization. We then categorize summarization approaches by the type of graphs taken as input and further organize each category by core methodology. Finally, we discuss applications of summarization on real-world graphs and conclude by describing some open problems in the field.

A spatially explicit reinforcement learning model for geographic knowledge graph summarization

“…Most graph summarization techniques fall into one of four categories (Liu et al., ): grouping‐ or aggregation‐based approaches; bit compression‐based approaches; simplification‐ or sparsification‐based approaches; and influence‐based approaches. Knowledge graph summarization usually adopts the simplification‐ or sparsification‐based approach because the prime motivation for summarizing knowledge graphs is to provide a subgraph that highlights the important entities and relations of the original graph.…”

“…Most graph summarization techniques fall into one of four categories (Liu et al, 2018): grouping-or aggregation-based approaches; bit compression-based approaches; simplification-or sparsification-based approaches;…”

“…Given the inherent richness of geospatial semantics (Yan, Janowicz, Mai, & Gao, ;Yan, Janowicz, Mai, & Zhu, ), geospatial components such as spatial contexts play a significant role in understanding spatial entities and their dependencies. However, existing (knowledge) graph summarization methods (Liu, Safavi, Dighe, & Koutra, ) are not tailored toward the geospatial domain, thus neglecting such special components. For instance, a summary about Santa Barbara, CA, is also always a partial summary of California.…”

Latent Network Summarization

“…Compared to the network embedding problem, latent network summarization differs in that it aims to derive a size-independent representation of the graph. This can be achieved in the form of supergraphs [19] (in the original graph space) or aggregated clusters trivially, but the compressed latent network summary in Definition 1 also needs to be able to derive the node embeddings, which is not the goal of traditional graph summarization methods.…”

“…This is in contrast to the goal of latent network summarization, which is to learn a size-independent representation of the graph. Latent network summarization also differs from traditional summarization approaches that typically derive supergraphs (e.g., mapping nodes to supernodes) [19], which target different applications and are unable to derive node embeddings.…”