基于深度学习的信息级联预测方法综述

Review of Information Cascade Prediction Methods Based on Deep Learning

在线社交媒体极大地促进了信息的产生和传递,加速了海量信息之间的传播与交互,使预测信息级联的重要性逐渐突显。近年来,深度学习已经被广泛用于信息级联预测(Information Cascade Prediction)领域。文中主要对基于深度学习的信息级联预测方法的研究现状与经典算法进行分类、梳理与总结。根据信息级联特征刻画的侧重点不同,将基于深度学习的信息级联预测方法分为时序信息级联预测方法与拓扑信息级联预测方法,并进一步将时序信息级联预测方法分为基于随机游走(Random Walk)的方法与基于扩散路径的方法,将拓扑信息级联预测方法分为基于全局拓扑结构的方法与基于邻域聚合的方法;并对每类方法进行详细的原理阐述与优缺点介绍,介绍了信息级联预测领域常用的数据集与评价指标,在宏观与微观两种信息级联预测场景下对基于深度学习的信息级联预测算法进行实验对比,并讨论了一些信息级联预测算法中常用的算法实现细节。最后,总结了该领域未来可能的研究方向与发展趋势。

Online social media greatly promotes the generation and transmission of information,exacerbates the communication and interaction between massive amounts of information,and highlights the importance of predicting information cascades.In recent years,deep learning has been widely used in the field of information cascade prediction.This paper mainly classifies,sorts,and summarizes the current research status of deep learning-based information cascade prediction methods and classic algorithms.According to the different emphasis of information cascade feature characterization,the information cascade prediction method based on deep learning is divided into time series information cascade prediction method and topology information cascade prediction method.The time series information cascade prediction method is further divided into methods based on random walks and methods based on diffusion paths,and the topology information cascade prediction method is divided into methods based on global topological structure and methods based on neighborhood aggregation.This paper details the principles and advantages and disadvantages of each type of method,and introduces the data sets and evaluation indicators commonly used in the field of information cascade prediction,and compares the information cascade prediction algorithms based on deep learning in the macro and micro information cascade prediction scenarios,and discusses some technical details commonly used in information cascade prediction algorithms.Finally,this paper summarizes the field possible future research directions and development trends.