Prediction For The Diffusion Of Behaviors Or Attitudes In Social Networks Based On Latent Space Models

Social network data represents the interactions and relationships between individuals in groups.One aspect of social interaction is social influence,that is,a person's attitude or behavior will change due to the social network he is on.The social network embodies two different aspects of social life:(1)a stable diffusion path,that is,something spreads through a connected group;(2)a random selection from a potential social space,which shows that people in the network the relative position between each other.This duality of the Internet brings challenges: Is it certain that attitudes or behaviors only spread along stable connections in social groups? This paper introduces the social space diffusion model to study the integration of these two viewpoints.In this model,the network relationship is represented by the position in the social space,and the diffusion in the social space is inversely proportional to the distance.The simulation process is divided into two parts.First,use the latent space model to estimate the individual's position in the latent space.Then,use the connection probabilities predicted from the model(derived from the distance between points in the latent space)or a series of networks randomly generated from these probabilities(representing regular changes in the network),and standardize the reciprocal of the distance between points in the latent space,as the weight in the weighted average framework.Using the longitudinal data of the friendship network of middle school students,the nature of the model is discussed.The results show that the smooth diffusion results produced by this model are 5% to 10% better than the diffusion model using the observation network to predict attitudes or behaviors in future surveys.