情報処理学会 第81回全国大会講演要旨

In recent years, the scale of E-commerce web sites and the number of users have increased remarkably. Along with that, it is required to respond promptly to requests from customers via online. In this research, we identify costumers, who want to withdraw their order by analyzing their e-mail text sent to customer support for classification model to urge operators to respond. The classification model was carried out in a convolution neural network after embedding expressions of words. We tuned the parameters in the experiment and found it possible to classify with accuracy of 79.2%. In addition, we compared the accuracy with predictive model by customer status. As a result, we obtained high accuracy compared with logistic regression models and support vector machines. From these results, it was proven that the mail document contained information indicating to resign from EC site customers.