Efficient Processing of Growing Temporal Graphs

Abstract

Temporal graphs are useful in modeling real-world networks. For example, in a phone call network, people may communicate with each other in multiple time periods, which can be modeled as multiple temporal edges. However, the size of real-world temporal graphs keeps increasing rapidly (e.g., considering the number of phone calls recorded each day), which makes it difficult to efficiently store and analyze the complete temporal graphs. We propose a new model, called equal-weight damped time window model, to efficiently manage temporal graphs. In this model, each time window is assigned a unified weight. This model is flexible as it allows users to control the tradeoff between the required storage space and the information loss. It also supports efficient maintenance of the windows as new data come in. We then discuss applications that use the model for analyzing temporal graphs. Our experiments demonstrated that we can handle massive temporal graphs efficiently with limited space.