影响力扩散概率模型及其用于意见领袖发现研究

摘要

作为意见领袖识别基础的影响力扩散模型IDM存在两个缺陷:(1)由回复链结构断层或者帖子内容间接传播引起的影响力传递中断；(2)由灌水所导致的虚假影响力传递.为解决上述问题,文中提出了一种新的影响力扩散概率模型IDPM,进而建立了网络意见领袖筛选模型.该模型在相同兴趣空间上定义单个关键词语传播概率影响力,在帖子影响力定义中引入了有效关键词语概念,避免了上述缺陷；同时,在用户影响力计算时给每个帖子一个影响因子,用以整合其它有用信息,使模型具有开放性和包容性特点.在2010年12月到2011年5月网易社会新闻版块评论数据上的实验表明,文中方法是有效的,其平均精确率相对IDM模型提高了59.8％.%There exist two defects in the influence diffusion model, which is a base for opinion leader identification. One is the influence diffusion break caused by the broken reply chain or indirect content diffusion, and the other is illusive influence diffusion caused by flooding posts. To solve the above problems, this paper presents a new Influence Diffusion Probability Model (IDPM) , and then builts a network opinion leader identification model. In which, the diffusion probability influence of the single term is defined in the same interesting space, and the concept of valid term in post influence evaluation is introduced. An impact fact to be added to influence calculation, which integrates other useful information, and this leads the model is open and inclusive. The experiments in the data collected from the NetEase social news section from December 2010 to May 2011, show that the method proposed in this paper is valid, and the average precision is more 59. 8% than that of IDM.