面向大数据的淘宝卖家信用度的影响因素分析

Taobao sellers credit evaluation based on mass data

消费者在电商平台上购买商品时,并不能获得关于消费品的所有信息,只能通过卖家信用、商品图片和购买评价等指标来判定所选的网店以及商品是否可靠.其中,卖家信用对于消费者的参考尤其重要.对卖家信用度建模能够在一定程度上保护交易双方的合法利益,提高交易的成功率.文章基于部分线性可加模型,结合社会资本数据(如新浪微博),对淘宝卖家信用度进行建模分析:(1)对数据进行相关分析、异常值剔除、多重共线性消除等预处理;(2)利用集群Lasso变量选择方法,识别出对卖家信誉有显著影响的因素;(3)对识别出来的因素与卖家信用做简单线性拟合,得出的结果与实际情况相违背,故又使用广义可加模型实现对卖家信誉的预测分析.该信用度模型能够很好地识别刷单卖家,帮助买家防范卖家的欺诈行为.

In this paper, we model the credit of Taobao seller based on the partially linear additive model and social communication data （such as Sina weibo）. To control the impact of ＂good evaluation＂, we do not use it in our model. First, we delete the noisy data and the related variables which result in multicollinearity. A natural choice is to use linear model to fit the data, however, we find that linear model is not adequate. Then we apply the partially linear additive model to analyze the data, and it indicates that this model performs better than traditional linear model.