A dual-layer dynamic graph summarization method based on extendable suffix fingerprints

The existing graph summarization methods often suffer from high addressing overheads and hash collisions, especially when facing real-world graph streams and power-law distributions, resulting in severe spatial-temporal performance degradation. This paper proposes a dual-layer dynamic graph summarization method (DLS). DLS is composed of inter-block and intra-block layers. In the inter-block layer, DLS employs an extendable suffix hash fingerprint-based addressing method, to achieve efficient inter-block addressing and migration. In the intra-block layer, DLS adopts a window-based adaptive extension mechanism, which adjusts the maximum extension size based on the degree statistics to reduce the intra-block hash collisions. The extensive experimental results on real-life graph datasets demonstrate DLS’s effectiveness. Compared with existing graph stream summarization methods, DLS can achieve an average of approximately 50 % performance promotion, 23 % average memory consumption saving than the traditional works.