Journal of KIISE (정보과학회 논문지)

Korean Institute of Information Scientists and Engineers (한국정보과학회)
권호 표지
DOI QR코드

DOI QR Code

Speakers' Intention Analysis Based on Partial Learning of a Shared Layer in a Convolutional Neural Network

Convolutional Neural Network에서 공유 계층의 부분 학습에 기반 한 화자 의도 분석

  • 김민경 (강원대학교 컴퓨터정보통신공학과) ;
  • 김학수 (강원대학교 컴퓨터정보통신공학과)
  • Received : 2017.05.22
  • Accepted : 2017.10.18
  • Published : 2017.12.15
⟨ Previous Next ⟩

Abstract

In dialogues, speakers' intentions can be represented by sets of an emotion, a speech act, and a predicator. Therefore, dialogue systems should capture and process these implied characteristics of utterances. Many previous studies have considered such determination as independent classification problems, but others have showed them to be associated with each other. In this paper, we propose an integrated model that simultaneously determines emotions, speech acts, and predicators using a convolution neural network. The proposed model consists of a particular abstraction layer, mutually independent informations of these characteristics are abstracted. In the shared abstraction layer, combinations of the independent information is abstracted. During training, errors of emotions, errors of speech acts, and errors of predicators are partially back-propagated through the layers. In the experiments, the proposed integrated model showed better performances (2%p in emotion determination, 11%p in speech act determination, and 3%p in predicator determination) than independent determination models.

대화에서 화자의 의도는 감정, 화행, 그리고 서술자로 표현될 수 있다. 따라서 사용자 질의에 정확하게 응답하기 위해서 대화 시스템은 발화에 내포된 감정, 화행, 그리고 서술자를 파악해야한다. 많은 이전 연구들은 감정, 화행, 서술자를 독립된 분류 문제로 다뤄왔다. 그러나 몇몇 연구에서는 감정, 화행, 서술자가 서로 연관되어 있음을 보였다. 본 논문에서는 Convolutional Neural Netowork를 이용하여 감정, 화행, 서술자를 동시에 분석하는 통합 모델을 제안한다. 제안 모델은 특정 추상화 계층과, 공유 추상화 계층으로 구성된다. 특정 추상화 계층에서는 감정, 화행, 서술자의 독립된 정보가 추출되고 공유 추상화 계층에서 독립된 정보들의 조합이 추상화된다. 학습 시 감정의 오류, 화행의 오류, 서술자의 오류는 부분적으로 역 전파 된다. 제안한 통합 모델은 실험에서 독립된 모델보다 좋은 성능(감정 +2%p, 화행 +11%p, 서술자 +3%)을 보였다.

Keywords

Acknowledgement

Supported by : 한국연구재단, 강원대학

References

  1. H. Kim, “Efficient Semantic Structure Analysis of Korean Dialogue Sentences Using an Active Learning Method,” Journal of KIISE: Software and Applications, Vol. 35, No. 5, pp. 306-312, 2008.
  2. C. Lee, S. Jung, J. Eun, M. Jeong, and G. Lee, "A situation-based dialoogue management using dialogue examples," Proc. of Acoustics, Speech and Signal Processing (ICASSP), IEEE International Conference, pp. 69-72, 2006.
  3. T. Mikolov, S. Kombrink, L. Burget, J. Cernocký, and S. Khudanpur, "Extensions of recurrent neural network language model," Proc. of Acoustics, Speech and Signal Processing (ICASSP), IEEE International Conference, pp. 5528-5531, 2011
  4. R. Collobert, J. Weston, L. Bottou, M. Karlen, K. Kavukcuoglu, and P. Kuksa, "Natural language processing (almost) from scratch," Journal of Machine Learning Research 12(Aug), pp. 2493-2537, 2011 12. Aug. (2011): 2493-2537.
  5. S. Joty and E. Hoque, "Speech act modeling of written asynchronous conversations with task-specific embeddings and conditional structured models," Proc. of the 54th Annual Meeting of the Association for Computational Linguistics, ACL, pp. 7-12, 2016.
  6. M. Kim and H. Kim, "User Utterance Analysis using Convolutional Neural Network," Journal of Telecommunications and information, Vol. 21, pp. 5-7, 2017.
  7. X. Wang, W. Jiang, and Z. Luo, "Combination of convolutional and recurrent neural network for sentiment analysis of short texts," Proc. of the 26th International Conference on Computational Linguistics, pp. 2428-2437, 2016.
  8. H. Lee, H. Kim, and J. Seo, J, “An Integrated Neural Network Model for Domain Action Determination in Goal-oriented Dialogues,” Journal of Information Processing Systems, Vol. 9, No. 2, pp. 259-270, 2013. https://doi.org/10.3745/JIPS.2013.9.2.259
  9. C. Seon, H. Lee, H. Kim, and J. Seo, "Improving Domain Action Classification in Goal-oriented Dialogues Using a Mutual Retraining Method," Journal of Pattern Recognition Letters, Vol. 45, pp. 154-160, 2014. https://doi.org/10.1016/j.patrec.2014.03.021
  10. Y. Goldberg and O. Levy, "Explained: deriving Mikolov et al.'s negative-sampling word-embedding method," arXiv preprint arXiv:11402.3722, 2014.