SNS中影响力用户预测研究——基于不平衡数据的多种数据挖掘方法对比

PREDICTING INFLUENTIAL USERS IN SNS:THE COMPARISON OF DATA MINING METHODS BASED ON IMBALANCED DATA

识别和预测SNS中某一话题中有影响力的用户,对于营销战略的精准实施和营销成本的节约具有重要意义.已有相关研究大都聚焦于静态情境下的用户影响力识别,而对于用户影响力预测问题的探讨还非常少.因此,本研究提出了一个综合的预测研究框架,结合不同的预测指标选取策略、属性选择方法和数据采样处理流程,并应用多种分类预测算法,对微博平台的用户影响力进行了实证分析.通过多种预测效果评价指标进行验证,表明本文提出的方法在用户影响力预测上具有较高的精度,显著提高了在个人层面的预测效果,在用户影响力预测方面具有较好的应用价值.

Identifying and predicting influential users in SNS under a certain topic is very important issue for marketing strategy implement and cost saving.The identification has been greatly focused on static context in available literatures,but the prediction is rarely researched.Hence,a synthesized prediction research framework is proposed in this paper to analyze the users＇ influence in SNS,in which different indicators＇ datasets,different feature selection methods,different proportions of data oversampling and multiple prediction algorithms are included.Our empirical result,which is validated by different performance evaluation indicators,shows that the proposed method is of high precision in predicting users＇ influence in SNS,and the prediction accuracy at individual level has been significantly improved.Therefore,the proposed method is good for practical application in predicting users＇ influence in SNS.