BFSlla

BFSlla is a graph traversal technique derived from the traditional breadth-first search (BFS). It introduces locality-aware layering to guide the expansion of nodes, aiming to reduce exploration of distant or irrelevant regions in large graphs while maintaining the distance guarantees of BFS on unweighted graphs. The method emphasizes exploring nodes that share a dense local neighborhood with the current frontier, based on a locality score computed from factors such as node degree, neighborhood overlap, and recent edge changes.

Origin and development: BFSlla was described in an academic publication series from the Center for Graph Theory and Networks in 2012, with subsequent implementations in open-source graph libraries. It was proposed as both a theoretical framework and a practical technique for improving performance on dynamic graphs where edges are added or removed over time.

Algorithm and properties: BFSlla maintains the standard BFS distance labels while selecting the next vertex from the frontier according to a locality score. A min-heap or bucketed priority queue can be used to retrieve the next node. In static graphs, BFSlla yields the same distance results as BFS; in dynamic graphs, it supports incremental updates with near-linear-time amortized performance. The algorithm can be combined with existing graph representations, and variants exist to trade off speed versus update stability.

Applications and variants: It has been discussed for social networks, routing, and large-scale bipartite graphs. Variants include BFSlla-Parallel for multi-core execution and BFSlla-Stream for graphs that evolve over time.