TY - JOUR
T1 - Learning influence probabilities in diffusion networks without timestamps
AU - Wang, Yuchen
AU - Wang, Huidi
AU - Gao, Chao
AU - Fan, Kefeng
AU - Cheng, Hailong
AU - Shen, Zhijie
AU - Wang, Zhen
AU - Perc, Matjaž
N1 - Publisher Copyright:
© 2025 Elsevier Inc.
PY - 2025/10/15
Y1 - 2025/10/15
N2 - Inferring information diffusion networks plays a crucial role in social network analysis and various applications. Existing methods often rely on the infection times of nodes in diffusion processes to uncover influence relationships. However, accurately monitoring real-time temporal information is challenging and resource-intensive. Additionally, some approaches that do not utilize infection timestamps fail to adequately capture the strength of influence relationships among nodes. To address these limitations, we propose a novel method called Learning Influence Probabilities in diffusion Networks without timestamps (LIPN). LIPN introduces an enhanced correlation metric to measure the relationship between node infections, which is utilized in the pre-pruning stage to mitigate the negative impact of redundant candidate edges during the inference process. LIPN constructs a likelihood function for the diffusion process by considering the infection probability between nodes. Furthermore, to enhance the reliability of the inferred results, LIPN incorporates an optimization strategy that combines an expectation maximization algorithm with a variant of the simulated annealing algorithm. The experimental results validate the effectiveness of LIPN in both synthetic networks and real-world networks, highlighting its potential for empowering social network analysis and applications.
AB - Inferring information diffusion networks plays a crucial role in social network analysis and various applications. Existing methods often rely on the infection times of nodes in diffusion processes to uncover influence relationships. However, accurately monitoring real-time temporal information is challenging and resource-intensive. Additionally, some approaches that do not utilize infection timestamps fail to adequately capture the strength of influence relationships among nodes. To address these limitations, we propose a novel method called Learning Influence Probabilities in diffusion Networks without timestamps (LIPN). LIPN introduces an enhanced correlation metric to measure the relationship between node infections, which is utilized in the pre-pruning stage to mitigate the negative impact of redundant candidate edges during the inference process. LIPN constructs a likelihood function for the diffusion process by considering the infection probability between nodes. Furthermore, to enhance the reliability of the inferred results, LIPN incorporates an optimization strategy that combines an expectation maximization algorithm with a variant of the simulated annealing algorithm. The experimental results validate the effectiveness of LIPN in both synthetic networks and real-world networks, highlighting its potential for empowering social network analysis and applications.
KW - Infection status
KW - Influence probabilities
KW - Information diffusion
KW - Simulated annealing algorithm
KW - Social networks
UR - http://www.scopus.com/inward/record.url?scp=105004265282&partnerID=8YFLogxK
U2 - 10.1016/j.amc.2025.129502
DO - 10.1016/j.amc.2025.129502
M3 - 文章
AN - SCOPUS:105004265282
SN - 0096-3003
VL - 503
JO - Applied Mathematics and Computation
JF - Applied Mathematics and Computation
M1 - 129502
ER -