TY - JOUR
T1 - FastGraph
T2 - Fast Large-Scale Directed Social Network Graph Generation
AU - Li, Qiyuan
AU - Liu, Ruoyi
AU - Chen, Renjiang
AU - Song, Tian
N1 - Publisher Copyright:
© 2014 IEEE.
PY - 2025
Y1 - 2025
N2 - Social networks are valuable tools for daily communication, information sharing, and news dissemination. For instance, on Twitter, users stay connected by following or mutually following each other. Prior works often study the characteristics of social networks using random graph models to capture the complexity of these relationships. To our knowledge, recent research not only generates directed social network graphs with reciprocal edges but also considers the rank correlations between different degree sequences. However, current generation methods are computationally expensive, particularly when applied to large-scale graphs. To address this, we propose a new graph generation model called FastGraph, designed to efficiently generate large-scale directed social network graphs with reciprocal edges and high clustering characteristics while significantly reducing computational costs. FastGraph introduces two key techniques: dynamic sampling based on current node degrees to reduce redundant computations and surplus degree rewiring to enhance local clustering. Compared to the latest Chung–Lu-based methods, FastGraph reduces the computational complexity from O(n3) to O(n2). When generating large-scale graphs with the same number of nodes and similar edge counts, FastGraph achieves an average runtime reduction of nearly 12 times. Furthermore, FastGraph successfully replicates more than a dozen social network graphs and is capable of efficiently generating graphs of arbitrary size while preserving properties closely aligned with those of real-world networks.
AB - Social networks are valuable tools for daily communication, information sharing, and news dissemination. For instance, on Twitter, users stay connected by following or mutually following each other. Prior works often study the characteristics of social networks using random graph models to capture the complexity of these relationships. To our knowledge, recent research not only generates directed social network graphs with reciprocal edges but also considers the rank correlations between different degree sequences. However, current generation methods are computationally expensive, particularly when applied to large-scale graphs. To address this, we propose a new graph generation model called FastGraph, designed to efficiently generate large-scale directed social network graphs with reciprocal edges and high clustering characteristics while significantly reducing computational costs. FastGraph introduces two key techniques: dynamic sampling based on current node degrees to reduce redundant computations and surplus degree rewiring to enhance local clustering. Compared to the latest Chung–Lu-based methods, FastGraph reduces the computational complexity from O(n3) to O(n2). When generating large-scale graphs with the same number of nodes and similar edge counts, FastGraph achieves an average runtime reduction of nearly 12 times. Furthermore, FastGraph successfully replicates more than a dozen social network graphs and is capable of efficiently generating graphs of arbitrary size while preserving properties closely aligned with those of real-world networks.
KW - Computational complexity
KW - high clustering
KW - random graph models
KW - social network generation
UR - https://www.scopus.com/pages/publications/105010889724
U2 - 10.1109/TCSS.2025.3582633
DO - 10.1109/TCSS.2025.3582633
M3 - Article
AN - SCOPUS:105010889724
SN - 2329-924X
VL - 12
SP - 4950
EP - 4964
JO - IEEE Transactions on Computational Social Systems
JF - IEEE Transactions on Computational Social Systems
IS - 6
ER -