Detecting Top-k Active Inter-Community Jumpers in Dynamic Information Networks

Abstract

Dynamic information networks, containing evolving objects and links, exist in various applications. Mining such networks is more challenging than mining static ones. In this paper, we propose a novel concept of Active Inter-Community Jumpers (AICJumpers) for dynamic information networks, which are objects changing communities frequently over time. Given communities of several snapshots in a dynamic network, we devise a time-efficiency top-k AICJumpers detection algorithm with a sliding window model. After denoting the jump score which captures how frequently an object changes communities over time, we encode the community changing trajectory of each object as bit vectors and transform jump scores computation into bitwise and, or and xor operations between bit vectors. We further propose a slide-based strategy for space and time saving. Experiments on both real and synthetic datasets show high effectiveness and efficiency of our methods as well as the significance of the AICJumper concept.