An Improved Model for Effective Customer Network Future State Prediction

• Published in: 2018 26th International Conference on Geoinformatics pp. 1 - 4
• Main Authors: Rui, Zeng, Hongli, Yin, Jinyan, Cai
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2018
• Subjects: Analytical models; anti-preferential attachment probability; Computational modeling; Data models; degree distribution; mean-field theory; Network topology; node deletion; power-law distribution; Predictive models; Simulation; Social networking (online); Time measurement; Time series analysis; Topology
• ISSN: 2161-0258
• DOI: 10.1109/GEOINFORMATICS.2018.8557185

For the purpose of prediction analysis of customer relationships in social networks, this paper proposes a simple model that can generate future states of a social network based on relevant data analysis and previous research. In this model, we insert nodes and edges of a customer network at the same preferential attachment probabilities, but delete them at different anti-preferential attachment probabilities with the consideration of the limit of network size, the directions of incident links and the factor of time in attractiveness. Furthermore, we propose an improved model based on the simple model that computes the attractiveness measure of nodes by applying time series prediction and takes into account of node in-degrees. Networks generated from our models have a nice property that their in-degree distribution follows the power-law, which desirably characterizes an essential property of social networks. This property is derived by applying the mean-field theory [7]. It is validated through simulation that this model can effectively generate a social network's future state and incorporating the factor of node in-degrees in computing the attractiveness measure of nodes using time series prediction can improve the prediction result.