文章摘要
汪加婧*,范维**.基于transformer的工单智能判责方法研究[J].高技术通讯(中文),2021,31(6):660~665
基于transformer的工单智能判责方法研究
The realization of intelligent judgments of the work order responsibilities based on transformer
  
DOI:10.3772/j.issn.1002-0470.2021.06.011
中文关键词: 工单智能判责; 文本分类; transformer; 双向编码器表示(BERT)
英文关键词: intelligent judgments of the work order responsibility, text categorization, transformer, bidirectional encoder representations from transformer(BERT)
基金项目:
作者单位
汪加婧*  
范维**  
摘要点击次数: 979
全文下载次数: 612
中文摘要:
      在图像、文本、视频、语音以及社交类网络数据爆炸增长的时代,企业如何从海量非结构化数据中提取出有效信息并将之转化为生产效率的提升和流程自动化的实现,是目前迫切需要关注和解决的问题。本文以运营商集团电子工单自动判责场景为切入点,提出使用基于transformer架构的双向编码器表示(BERT)作为文本分类模型,自动收集各省份的反馈信息并进行各省份的工单责任智能判定。通过将BERT模型与LightGBM和Bi-LSTM-Attention模型进行实验对比,结果表明BERT模型对各类别工单的预测准确率均达到了96%以上,具有较好的实际应用效果。
英文摘要:
      In the era of explosive growth of images, texts, videos, voice and social network data, how enterprises extract and transform effective information from massive unstructured data to improve the production efficiency and realize the process automation has been an urgent concern that needs to be solved. Taking the electronic work order automatic responsibility judgment scenario of the Telecom Operator Group as the starting point, the bidectional encoder representations from transformer (BERT) based on the transformer architecture is used as the text categorization model to automatically collect feedback information from provinces and make intelligent judgments of the work order responsibilities of each province. The comparison and analysis of the BERT model, the LightGBM model and the Bi-LSTM-Attention model indicate that the prediction accuracy of the BERT model on all types of work orders is over 96%, showing excellent practical effects.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮