基于贝叶斯收缩估计的有影响力消费者识别

Identifying influential consumers based on Bayesian shrinkage

在应用网络科学方法寻找有影响力的消费者时,大多数以往研究存在着一个重要假设:消费者之间的相互影响为正,或无影响,并未考虑到消费者之间可能存在的负向影响.鉴于此,本文放松以往的假设,把消费者之间的相互作用设定为正向,负向以及无影响.为了验证以上假设,本文以虚拟世界中4429名消费者之间构建的186253条朋友关系链为样本,应用泊松回归来估计消费者之间的相互影响.为了提升估计的效率,本文提出了一种基于贝叶斯收缩的新估计方法.研究结果表明:相比于以往假设提出的模型,基于本文假设提出的模型能够更好地拟合数据.本文的研究不仅从理论上验证了消费者之间负向影响的存在,也为企业寻找有影响力的消费者提供了一种更为精确的方法.

When identifying influential consumers via network science analysis,previous scholars held an important assumption that all users produce positive(or no)influence on others,without considering the potential negative effects among users.Therefore,this paper loosens previous assumption and sets consumers’influence as three type:Positive,none or negative.To verified our assumption and probe influence among users,we adopt Poisson regression and use data from virtual world,which including 4,429 consumers and 186,253 friend links.Meanwhile,we proposed a new estimation method via Bayesian shrinkage to elevate estimation efficiency.The results show that the model based on our assumption has a better fitness than the models based on previous assumptions.Our research not only verify the possible negative influence among users,but also provide a more accurate methodology to find influential users from network science perspective.