Journal of KIISE (정보과학회 논문지)

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

DOI QR Code

Categorization of Korean News Articles Based on Convolutional Neural Network Using Doc2Vec and Word2Vec

Doc2Vec과 Word2Vec을 활용한 Convolutional Neural Network 기반 한국어 신문 기사 분류

  • 김도우 (서강대학교 정보통신대학원) ;
  • 구명완 (서강대학교 컴퓨터공학)
  • Received : 2017.02.07
  • Accepted : 2017.05.16
  • Published : 2017.07.15
⟨ Previous Next ⟩

Abstract

In this paper, we propose a novel approach to improve the performance of the Convolutional Neural Network(CNN) word embedding model on top of word2vec with the result of performing like doc2vec in conducting a document classification task. The Word Piece Model(WPM) is empirically proven to outperform other tokenization methods such as the phrase unit, a part-of-speech tagger with substantial experimental evidence (classification rate: 79.5%). Further, we conducted an experiment to classify ten categories of news articles written in Korean by feeding words and document vectors generated by an application of WPM to the baseline and the proposed model. From the results of the experiment, we report the model we proposed showed a higher classification rate (89.88%) than its counterpart model (86.89%), achieving a 22.80% improvement. Throughout this research, it is demonstrated that applying doc2vec in the document classification task yields more effective results because doc2vec generates similar document vector representation for documents belonging to the same category.

본 논문에서는 문장의 분류에 있어 성능이 입증된 word2vec을 활용한 Convolutional Neural Network(CNN) 모델을 기반으로 하여 문서 분류에 적용 시 성능을 향상시키기 위해 doc2vec을 함께 CNN에 적용하고 기반 모델의 구조를 개선한 문서 분류 방안을 제안한다. 먼저 토큰화 방법을 선정하기 위한 초보적인 실험을 통하여, 어절 단위, 형태소 분석, Word Piece Model(WPM) 적용의 3가지 방법 중 WPM이 분류율 79.5%를 산출하여 문서 분류에 유용함을 실증적으로 확인하였다. 다음으로 WPM을 활용하여 생성한 단어 및 문서의 벡터 표현을 기반 모델과 제안 모델에 입력하여 범주 10개의 한국어 신문 기사 분류에 적용한 실험을 수행하였다. 실험 결과, 제안 모델이 분류율 89.88%를 산출하여 기반 모델의 분류율 86.89%보다 2.99% 향상되고 22.80%의 개선 효과를 보였다. 본 연구를 통하여, doc2vec이 동일한 범주에 속한 문서들에 대하여 유사한 문서 벡터 표현을 생성하기 때문에 문서의 분류에 doc2vec을 함께 활용하는 것이 효과적임을 검증하였다.

Keywords

Acknowledgement

Supported by : 국가과학기술연구회

References

  1. K.H.Joo, E.Y.Shin, J.I.Lee, W.S.Lee, "Hierarchical Automatic Classification of News Articles based on Association Rules," Journal of Korea Multimedia Society, Vol. 14, No. 6, pp. 730-741, Jun. 2011. (in Korean) https://doi.org/10.9717/kmms.2011.14.6.730
  2. Y.G.Beak, "A Study on Automatic Classification System of Hangul Internet News Articles," Korea University, Graduate School of Department of Business Administration, a master's thesis, 2003. (in Korean)
  3. Tomas Mikolov, Kai Chen, Greg Corrado, Jeffrey Dean, "Efficient Estimation of word Representations in Vector Space," arXiv:1301.3781v3, Sep. 2013.
  4. Yoon Kim, "Convolutional Neural Network for Sentence Classification," Proc. of the 2014 Conference on Empirical Methods in Natural Language Processing(EMNLP), pp. 1746-1751, Oct. 2014.
  5. D.W.Kim, M.W.Koo, "A Study on Categorization of Korean News Article based on CNN using Doc 2Vec," the 28th Annual Conference on Human and Cognitive Language Technology, pp. 67-71, 2016. (in Korean)
  6. Ye Zhang, Byron C. Wallace, "A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification," arXiv:1510.03820v4, Apr. 2015.
  7. Quoc Le, Tomas Milokov, "Distributed Representations of Sentences and Documents," Proc. of the 31st International Conference on Machine Learning, 2014.
  8. scikit-learn developers, "2.5. Decomposing signals in components (matrix factorization problems) scikitlearn 0.18 documentation," [Online]. Available:http://scikit-learn.org/stable/modules/decomposition.html (downloaded 2016, Oct. 28)
  9. M.J.Choi, H.S.Park, T.S.Jeong, S.H.Jeong, Y.H, S.H.Lee, Y.J.Hwang, "Understanding of NewsML (policy data report 2007-01)," Korea Press Foundation, pp. 10-18, 2007. (in Korean)
  10. E.J.Park, S.Z.Cho, "KoNLPy: Korean natural language processing in Python," the 26th Annual Conference on Human and Cognitive Language Technology, 2014. (in Korean)
  11. Mike Schuster and Kaisuke Nakajima, "JAPANESE AND KOREAN VOICE SEARCH," Google Inc, USA, 2012.
  12. J.H.Park, M.W.Koo, "A Study on the Sentiment analysis Google Play Store App Comment Based on WPM(Word Piece Model)," the 28th Annual Conference on Human and Cognitive Language Technology, pp. 291-295, 2016. (in Korean)