计算机工程 ›› 2019, Vol. 45 ›› Issue (7): 237-241.doi: 10.19678/j.issn.1000-3428.0051291

• 人工智能及识别技术 • 上一篇    下一篇

基于标签分解的口语理解模型

许莹莹, 黄浩   

  1. 新疆大学 信息科学与工程学院, 乌鲁木齐 830046
  • 收稿日期:2018-04-23 修回日期:2018-05-24 出版日期:2019-07-15 发布日期:2019-07-23
  • 作者简介:许莹莹(1990-),女,硕士研究生,主研方向为自然语言处理、语音通信;黄浩,教授、博士。
  • 基金项目:
    国家自然科学基金(61663044,61365005)。

Spoken Language Understanding Model Based on Label Decomposition

XU Yingying, HUANG Hao   

  1. College of Information Science and Engineering, Xinjiang University, Urumqi 830046, China
  • Received:2018-04-23 Revised:2018-05-24 Online:2019-07-15 Published:2019-07-23

摘要: 在双向长短时记忆网络的基础上,提出一种用于口语理解的标签拆分策略,并构建一个联合模型。通过将1次127种标签分类转换成3次独立的分类,平衡ATIS数据集的标签。针对ATIS数据集资源较少的问题,引入外部词向量以提升模型的分类性能。实验结果表明,与循环神经网络及其变体结构相比,该模型的F1值有显著提升,最高可达95.63%。

关键词: 口语理解, 槽填充, 双向长短时记忆网络, 词向量, 联合模型

Abstract: Based on the Bi-Long Short Term Memory (BiLSTM),this paper proposes a label splitting strategy for Spoken Language Understanding(SLU)and constructs a joint model.The model convert a classification of 127 labels into 3 independent classifications to balance the labels in the ATIS database.Due to the scarcity of ATIS data,this paper introduces external word embedding to improve the classification performance of the model.Experimental results show that compared with the traditional recurrent neural network and its variants,the proposed joint model obtains significantly improvement in F1 value,which can be up to 95.63%.

Key words: Spoken Language Understanding(SLU), slot filling, Bi-Long Short Term Memory(BiLSTM), word embedding, hibrid model

中图分类号: