Performing Distance Queries on Social Networks in Sublinear Time

Abstract

Shortest path computation is an important base task in many applications. While there have been improvements to the shortest path algorithms, all require preprocessing the entirety of the graph, creating inefficiencies, especially when applied to large social networks. Considering that social networks often appear with power law distributions, we present the question of utilizing this insight for sublinearity. We thus propose Wormhole, an algorithm that can perform reasonably accurate shortest distance estimations in sublinear runtime. On large graphs, scaling up to billions of edges, Wormhole empirically demonstrates the ability to provide reasonable accuracy over 10,000 distance queries while only seeing 𝑂( √ 𝑛) vertices. This shows an improvement over the baseline method of Bi-directional BFS, which has shown similar results on the scale of 𝑂(𝑛).