Predicting Graph Operator Output over Multiple Graphs

Abstract

A growing list of domains, in the forefront of which are Web data and applications, are modeled by graph representations. In content-driven graph analytics, knowledge must be extracted from large numbers of available data graphs. As the number of datasets (a different type of volume) can reach immense sizes, a thorough evaluation of each input is prohibitively expensive. To date, there exists no efficient method to quantify the impact of numerous available datasets over different graph analytics tasks. To address this challenge, we propose an efficient graph operator modeling methodology. Our novel, operator-agnostic approach focuses on the inputs themselves, utilizing graph similarity to infer knowledge about them. An operator is executed for a small subset of the available inputs and its behavior is modeled for the rest of the graphs utilizing machine learning. We propose a family of similarity measures based on the degree distribution that prove capable of producing high quality models for many popular graph tasks, even compared to modern, state of the art similarity functions. Our evaluation over both real-world and synthetic graph datasets indicates that our method achieves extremely accurate modeling of many commonly encountered operators, managing massive speedups over a brute-force alternative.