An Efficient Lossless Graph Summarization Method for Large Streaming Graphs

Abstract

Graph summarization aims to extract critical information from large graphs by creating summaries that represent the original data. Especially, real-world daily applications generate massive dynamic streaming graphs, representing as edge or node streams. How to efficiently generate compact and lossless graph summaries for large streaming graphs is still a challenging problem. This paper proposes an efficient and scalable lossless summarization method for streaming graphs, called Partition and Similarity-based Two-Stage Summarization (PSTSS). PSTSS uses a streaming graph partitioning algorithm as the first stage to generate coarse-grained summaries by on-the-fly subgraph partitioning. The second stage generates fine-grained supernodes within these coarse-grained supernodes through efficient node similarity calculation. Our experiments on six large datasets demonstrate that our method achieves better operational efficiency, compression rate, scalability, and graph query speed compared to state-of-the-art lossless summarization algorithms.