摘要
在社会网络中,量化分析用户影响力及用户之间的影响关系已经变得越来越有必要,对精度的要求也越来越高.网络模型构建的好坏,模型预测用户行为的好坏对商业营销、社会影响最大化等应用都有着至关重要的意义.如何从不同方面量化社会影响力?如何量化不同方面下的用户之间的影响力强弱?如何在社会网络中构造这样一个模型?为了解决这些问题,提出一个基于主题概率的从众性模型(Topical Conformity Model,简称TCM),模型从主题层对社会网络进行建模,考虑不同主题概率下的用户影响力以及用户之间的影响关系.将该模型应用在学术网络中的用户关键词预测上,并与前人的方法进行实验对比,各项预测指标都有一定的提升,其中AUC值提高了4.3%,验证了本文提出的模型对于用户行为预测的有效性.此外,本文的工作还解决了如下两个问题:1、寻找某个主题下最可能选择某行为的用户;2、寻找某个用户选择某个行为受影响最大的主题.
Quantitatively analyzing the influence and conformity of users has become increasingly necessary, as well as the high performance of prediction in social network. Constructing network model and predicting users＇ behavior become more and more critical for applications such as viral marketing, social influence maximization, etc. How to quantify social influence from different aspects？ How to quantify the strength of influence between users in different view？ How to build such a model in social network？ To solve these problem, we proposed a Topical Conformity Model （ TCM ）, modeling social network at the topic level and taking the influence and conformity of users under different topic into account. We apply the model to several academic research network, predicting word usage. Compared to baseline models, our model performs better, where AUC value increases 4.3 %, verifying that the model proposed in this paper is effective to word usage prediction. Furthermore, our work also solves the following two problems：1, finding users who are most likely to take specific action in a topic ; 2, finding most likely topics that result in user＇ s behavior.
出处
《小型微型计算机系统》
CSCD
北大核心
2017年第2期277-281,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61502170)资助
关键词
社会网络
影响从众性
主题概率
机器学习
social network
influence conformity
topic probability
machine learning
