GenCAT: Generating Attributed Graphs with Controlled Relationships between Classes, Attributes, and Topology

Abstract

Generating large synthetic attributed graphs with node labels is an important task to support various experimental studies for graph analysis methods. Existing graph generators fail to simultaneously simulate the relationships between labels, attributes, and topology which real-world graphs exhibit. Motivated by this limitation, we propose GenCAT, an attributed graph generator for controlling those relationships, which has the following advantages. (i) GenCAT generates graphs with user-specified node degrees and flexibly controls the relationship between nodes and labels by incorporating the connection proportion for each node to classes. (ii) Generated attribute values follow user-specified distributions, and users can flexibly control the correlation between the attributes and labels. (iii) Graph generation scales linearly to the number of edges. GenCAT is the first generator to support all three of these practical features. Through extensive experiments, we demonstrate that GenCAT can efficiently generate high-quality complex attributed graphs with user-controlled relationships between labels, attributes, and topology.

Attributed Labeled BTER-Based Generative Model for Benchmarking of Graph Neural Networks

“…Possibility to specify desired degree distribution and especially power-law degree distribution 3. Possibility to control properties of graph with input parameters We have chosen three generators to compare with: LFR [21], ADC-SBM [20], Gen-CAT [22] as they satisfy all requirements and significantly differ from each other in the mechanism of edge assignment. Further, we will consider only the case of undirected graphs.…”

Beyond Real-world Benchmark Datasets: An Empirical Study of Node Classification with GNNs

“…We adopt GenCAT graph generator [20], the only method satisfying the above two requirements. Current state-of-the-art methods [1,28] fail to support either of the requirements satisfied by GenCAT.…”

“…Another example is SBM [1] which does not generate graphs similar to real-world graphs as shown in [20] and does not support attribute generation, i.e., it fails to capture the third characteristics.…”

“…Also, GenCAT can simulate the existing generators SBM without any modifications of GenCAT itself. Please see more detailed and precise procedures in [20].…”

“…To understand the pros/cons of GNNs, we empirically study the performance of GNNs through extensive experiments on various graphs by synthetically changing one or a few target characteristic(s) of graphs. To this end, we adapt a recent graph generator [20] that can more flexibly control the characteristics of generated graphs with node labels than a traditional generator SBM. In terms of the above four major characteristics, we attempt to answer the following questions: Q1.…”