APMSID: Activated Probability for Multi-Source Information Diffusion in Online Social Networks

• Published in: IEEE access 2018, Vol.6, p.64435-64449
• Main Authors: Tan, Zhenhua, Wu, Danke, Yang, Guangming, Bin, Zhang
• Format: Article
• Language: English
• Subjects: activated probability; Algorithms; Analytical models; Diffusion; Homology; influence probability; influence propagation; information diffusion; Information dissemination; Integrated circuit modeling; Ions; Model accuracy; Peer-to-peer computing; Social network services; Social networks; Statistical analysis; Technological innovation
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2877797

Multi-source diffusion is a common phenomenon in online social networks (OSNs) that involves pieces of information with different purposes concurrently propagating over networks in a cooperative or competitive way. A traditional activated probability model for single-source diffusion cannot be directly applied to multi-source diffusion, especially when these diffusions interfere with each other. We consider that herd behavior exists over multi-source information diffusion in OSNs, and we propose a more rational activated probability pattern for multi-source diffusion, activated probability for multi-source information diffusion (APMSID), inspired by the herding effect. APMSID consists of three components to calculate the final activated probability to the destination node, including the influence of percentage (IOP), the influence of energy (IOE), and the influence of competition (IOC). APMSID fully considers the homologous diffusion (cooperation) personalization of the infected percentage by IOP and influence energy direction by IOE and considers the negative influence of IOC from competitive information diffusion. Mathematical modeling and algorithms are designed in detail. Experiments based on the real datasets of Epinions, wiki-Vote, Slashdot, and Micro-blog show that the proposed APMSID is more rational and accurate than the traditional method. APMSID can effectively solve the probabilistic reasoning problem of the multi-source diffusion phenomenon and is superior to the IC model in predicting the accuracy of the infected nodes.