计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (21): 188-194.DOI: 10.3778/j.issn.1002-8331.2102-0235

• 模式识别与人工智能 • 上一篇    下一篇

结合胶囊网络的领域适应意图识别

赵鹏飞,李艳玲,林民   

  1. 内蒙古师范大学 计算机科学技术学院,呼和浩特 010022
  • 出版日期:2021-11-01 发布日期:2021-11-04

Intent Detection of Domain Adaptation Combined with Capsule Network

ZHAO Pengfei, LI Yanling, LIN Min   

  1. College of Computer Science and Technology, Inner Mongolia Normal University, Hohhot 010022, China
  • Online:2021-11-01 Published:2021-11-04

摘要:

意图识别是口语理解中的重要任务,关乎整个对话系统的性能。针对新领域人机对话系统中训练语料较少,构建可训练语料十分昂贵的问题,提出一种利用胶囊网络改进领域判别器的领域适应方法。该方法利用领域对抗神经网络将源域的特征信息迁移至目标域中,此外,为了保证领域意图文本的特征质量,对源域和目标域的特征表示进行再次提取,充分获取意图文本的特征信息,捕捉不同领域的独有特征,提高领域的判别能力,保障领域适应任务的可靠性。在目标域仅包含少量样本的情况下,该方法在中文和英文数据集上的准确率分别达到了83.3%和88.9%。

关键词: 意图识别, 对话系统, 胶囊网络, 领域适应

Abstract:

Intent detection is an important task in spoken language understanding, which is related to the performance of the entire dialogue system. Aiming at the problem of less training corpus in the human-machine dialogue system in the new domain, and the construction of training corpus is very expensive. This thesis proposes a domain adaptation method using capsule network to improve the domain discriminator. This method uses a domain adversarial neural network to transfer the feature information of the source domain to the target domain, in addition, in order to ensure the feature quality of the domain intent text, the feature representations of the source domain and the target domain are extracted again, which can fully obtain the feature information of the intent text, captures the unique features of different domains, improves the discriminator ability of the domain, and ensures the reliability of the guarantee domain adaptation tasks. When the target domain contains only a small number of labeled samples, the accuracy rate on Chinese and English datasets reaches 83.3% and 88.9%.

Key words: intent detection, dialogue system, capsule network, domain adaptation