Learning Asynchronous-Time Information Diffusion Models and its Application to Behavioral Data Analysis over Social Networks

摘要

One of the interesting and important problems of information diffusion over alarge social network is to identify an appropriate model from a limited amountof diffusion information. There are two contrasting approaches to modelinformation diffusion: a push type model known as Independent Cascade (IC)model and a pull type model known as Linear Threshold (LT) model. We extendthese two models (called AsIC and AsLT in this paper) to incorporateasynchronous time delay and investigate 1) how they differ from or similar toeach other in terms of information diffusion, 2) whether the model itself islearnable or not from the observed information diffusion data, and 3) whichmodel is more appropriate to explain for a particular topic (information) todiffuse/propagate. We first show there can be variations with respect to howthe time delay is modeled, and derive the likelihood of the observed data beinggenerated for each model. Using one particular time delay model, we show themodel parameters are learnable from a limited amount of observation. We thenpropose a method based on predictive accuracy by which to select a model whichbetter explains the observed data. Extensive evaluations were performed. Wefirst show using synthetic data with the network structures taken from realnetworks that there are considerable behavioral differences between the AsICand the AsLT models, the proposed methods accurately and stably learn the modelparameters, and identify the correct diffusion model from a limited amount ofobservation data. We next apply these methods to behavioral analysis of topicpropagation using the real blog propagation data, and show there is a clearindication as to which topic better follows which model although the resultsare rather insensitive to the model selected at the level of discussing how farand fast each topic propagates from the learned parameter values.