电信科学 ›› 2021, Vol. 37 ›› Issue (6): 125-131.doi: 10.11959/j.issn.1000-0801.2021125

• 研究与开发 • 上一篇    

基于GRW和FastText模型的电信用户投诉文本分类应用

赵进, 杨小军   

  1. 中国电信股份有限公司重庆分公司,重庆 401120
  • 修回日期:2021-05-25 出版日期:2021-06-01 发布日期:2021-06-01
  • 作者简介:赵进(1989- ),男,现就职于中国电信股份有限公司重庆分公司,主要研究方向为大数据、人工智能、开源软件等
    杨小军(1970- ),男,现就职于中国电信股份有限公司重庆分公司,主要研究方向为云计算、大数据、人工智能、开源软件等

Application on text classification of telecom user complaints based on GRW and FastText model

Jin ZHAO, Xiaojun YANG   

  1. Chongqing Branch of China Telecom Co., Ltd., Chongqing 401120, China
  • Revised:2021-05-25 Online:2021-06-01 Published:2021-06-01

摘要:

随着神经网络的广泛应用,将神经网络应用到自然语言处理文本分类问题中,成为一种有效的解决方法。电信运营商客户服务中心通过多种渠道收集用户投诉信息,为了对投诉文本信息进行自动分类并将其落实到具体责任部门,提升用户感知,提出了一种基于GRW模型和FastText模型的文本分类方法。首先通过GRW模型对投诉文本进行特征选择,提取有效特征词;然后构建基于FastText模型的用户投诉文本分类方法;最后在公开数据集和运营商已标注的投诉文本数据集上进行实验。结果表明,基于 GRW 和 FastText模型的文本分类方法比朴素贝叶斯、双向LSTM和Bert模型在准确率、Kappa系数及汉明损失方面的性能有较大提升。

关键词: 神经网络, 文本分类, GRW模型, FastText模型

Abstract:

With the widespread application of neural network, the application of neural network to natural language processing text classification problems has become an effective solution.The customer service center of telecom operator collected user complaint information from multiple channels.In order to automatically classify the complaint text information and assign it to the specific responsible department for processing and reply, enhancing customer perception further, a textclassification method based on GRW and FastTextmodel was proposed.Firstly, the GRW model was used to select the features of the complaint text, extract effective feature words, and then a user complaint text classification method based on FastText model was constructed.Experiments on public datasets and a complaint text data by one of telecom company show that the text classification method based on GRW and FastText model is better than naive Bayes, bidirectional LSTM and Bert pre-trained model in accuracy, Kappa coefficient and Hamming loss.

Key words: neural network, text classification, GRW model, FastText model

中图分类号: 

No Suggested Reading articles found!