基于特征提取与分层建模的社交网络信息传播预测研究

The problem of propagation is an important research topic in the area of complex networks. Predicting the propagation of information has significant guiding value to opinion controlling, advertising and a series of application fields of social media. Our work is based on the method of complex networks and statistical learning theory. The object of our study is to find out the potential pattern of information spreading, and to deduce the possibility of spreading the information by a node, and then to predict the procedure and result of a propagation process through the dynamical model that would be made..The details are as follows: Firstly, multilayer model will be built based on a feature of the node to analyze the rule of how information flow within and across the layers；Then to find a way measuring the importance of a node for the spreading of a particular information and figure out the probability of a node to spread the message; Finally, a model of propagation dynamics will be designed to simulate the process of spreading, which can help to forecast the possible results of the propagation..Our research is based on the cutting-edge theory of complex networks, with the support of the large scale data of social networks. It is also a joint work of information collection and machine learning. So the results of this work would provide an important reference to the solution of lots of practical application fields.

传播问题是复杂网络理论的重要研究课题之一，其中关于信息传播预测的研究成果，在舆情控制、广告宣传等众多社会媒体应用领域越来越显示出其指导意义。本项目以在线社交网络的信息传播为研究对象，结合复杂网络分析方法和统计学习方法，旨在分析潜在的传播规律，推断节点产生传播行为的可能性，并进一步通过传播动力学模型预测传播的规模和结果。具体的研究内容包括：建立基于节点特征的社交网络分层模型，并分析信息在不同特征层及层间的传播规律；提出基于特定内容的节点特征重要性度量方法，继而建立计算节点传播概率的算法；结合分析和计算的结果，设计信息传播的动力学模型，从而对社交网络上的传播行为进行模拟，最终实现对整个传播过程和结果的预测。本项目以复杂网络的前沿理论为基础，社交网络大规模数据为支撑，结合信息采集技术和机器学习算法，展开多学科交叉研究，研究成果将为准确预测信息传播，指导实际的相关问题提供重要参考。

本项目基于微博社交网络大数据，旨在分析微博社交网络结构及传播动力学，并最终实现对传播进行一定程度的预测。项目经过三年的执行，完成了立项时的各项任务，实现了基于特征的分层分析、基于微博内容的传播分析、以及基于机器学习算法的传播动力学建模等创新性的研究，并在节点重要性分析、社团划分、关键词提取、情感分析以及传播动力学建模等领域取得了相应成果。. 项目进行过程中项目团队共发表SCI论文7篇，会议论文1篇，北大中文核心3篇，其他论文2篇，申请专利19项，培养硕士研究生9名，形成基于微博分析的舆情应用平台2个，完成了预期的研究目标。本项目的研究案例支撑了西南石油大学数据科学与大数据技术专业的部分课程。同时，本项目成果支撑了项目团队的“舆情事件分析平台”“高校舆情平台”等舆情分析平台的传播分析部分，这两个平台解决了相关部门的舆情分析和精准营销等实际问题，具有较大的实际效益。