How to make an accurate evaluation of the quality of pension service has become the most important task. However, in the real world, many customs always forget to rate pension service. They only leave a few short, less semantic, and discontinuous review words below the service. This paper will propose an effective multi-dimension attention convolutional neural networks (MACNNs) model to analyze customer review texts and predict the pension service quality. In MACNN, the emoticon feature, sentiment feature, and word feature can be extracted together to construct feature space. And then attention layer and convolution layer work together to predict the service quality. Compared with the traditional machine learning methods and neural network methods, this method is more objective and accurate to reflect consumers’ real evaluation of pension service.

中文翻译：

---

一种利用多维注意力卷积神经网络评估养老金服务质量的有效方法

如何对养老服务质量做出准确的评价已成为当务之急。然而，在现实世界中，许多海关总是忘记对养老金服务进行评级。他们只在服务下方留下一些简短、语义较少且不连续的评论词。本文将提出一种有效的多维注意力卷积神经网络 (MACNNs) 模型来分析客户评论文本并预测养老金服务质量。在 MACNN 中，可以将表情特征、情感特征和单词特征一起提取来构建特征空间。然后注意力层和卷积层协同工作来预测服务质量。与传统的机器学习方法和神经网络方法相比，该方法更加客观准确地反映了消费者对养老服务的真实评价。