Metin Temsil Yöntemlerine Yönelik Farklı Yakla¸sımların Kar¸sıla¸stırılması

Son zamanlarda derin ö˘grenme mimarileri bir çok do˘gal dil i¸sleme problemini ba¸sarılı bir ¸sekilde çözmü¸stür. Açık kaynak kodlu kütüphanelerin yaygınlı˘gı yapay zeka yakla¸sımlarını daha uygulanabilir hale getirmi¸stir. Teknolojideki bu ani ivmelenme do˘gal dil i¸slemedeki standartları dönü¸stürdü ve geli¸stirdi. Bu çalı¸smada açık kaynak kodların ve alanla ilgili ara¸stırmaların rahat eri¸sebilirli˘gi sayesinde metin temsiliyeti yakla¸sımlarının önemli bir kısmını de˘gerlendirme imkanı bulduk. Dört farklı paradigmayı metin temsiliyeti açısından de˘gerlendirdik: Geleneksel kelime torbası yakla¸sımı, konu modelleme, gömme temsiliyeti ve derin ö˘grenme. Çalı¸smanın ana katkısı olarak, Türkçe için metin sınıflandırma problemini tüm bu metin temsiliyetlerini ve ilgili makine ö˘grenme algoritmalarını kullanarak ele aldık. Olu¸sturulan denetimli ö˘grenme mimarisi özellikle Türkçe için hazırlanmı¸s bir veri seti ile sınanmı¸stır. Her bir temsiliyet için onunla uyumlu çalı¸sacak SVM, çok-katlı Naive Bayes (mNB) gibi makine ö˘grenmesi algoritmaları sınandı. Çe¸sitli deneyler sonucunda ba¸sarılı bir metin sınıflandırıcı mimarisinin Türkçe için nasıl kurulaca˘gını bu makalede tartı¸stık ve ba¸sarılı modeller sunduk. Son olarak kelime torbası gibi geleneksel yöntemlerin hala ba¸sarılı oldu˘gunu ve derin ö˘grenme temelli modellerin bazılarından daha iyi oldu˘gunu gördük.

A Comparison of Different Approaches to Document Representation in Turkish LanguageA Comparison of Different Approaches to Document Representation in Turkish Language

Recently, deep learning methods have demonstrated state-of-the-art performancein numerous complex Natural Language Processing (NLP) problems. Easy accessibilityof high-performance computing resources and open-source libraries makes ArtificialIntelligence (AI) approaches more applicable for researchers. This sudden growth ofavailable techniques shaped and improved standards in the field of NLP. Thus, we find anopportunity to compare different approaches to document representation, owing to variousopen-source libraries and a large amount of research. We evaluate four different paradigmsto represent documents: Traditional bag-of-words approaches, topic modeling, embeddingbased approach and deep learning. As the main contribution of this article, we aim atevaluating all these representation approaches with suitable machine learning algorithmsfor document categorization problem in the Turkish language. The supervised architectureuses a benchmark dataset specifically prepared for this language. Within the architecture,we evaluate the representation approaches with corresponding machine learning algorithmssuch as Support Vector Machine (SVM), multi-nominal Naive Bayes Algorithm(m-NB) and so forth. We conduct a variety of experiments and present successful resultsfor the Turkish document categorization. We also observed that tradition approaches havestill comparable results with Neural Network models in terms of document classification.

PDF

___

B. Açıkalın and N. G. Bayazıt. 2016. The importance of preprocessing in Turkish Text classification. In 2016 24th Signal Processing and Communication Application Conference (SIU). 2053–2056. https: //doi.org/10.1109/SIU.2016.7496174
Burak Kerim Akkus and Ruket Çakıcı. 2013. Categorization of Turkish News Documents with Morphological Analysis. (2013).
Mehmet Fatih Amasyalı, Sümeyra Balcı, Emrah Mete, and Esra Nur Varlı. 2012. A Comparison of Text Representation Methods for Turkish Text Classification. EMO Scientific Journal 2 (2012). arXiv:1309-5501
M. Fatih Amasyalı and Banu Diri. 2006. Automatic Turkish Text Categorization in Terms of Author, Genre and Gender. In Natural Language Processing and Information Systems, Christian Kop, Günther Fliedl, Heinrich C. Mayr, and Elisabeth Métais (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 221– 226.
Ça˘grı Toraman. 2011. Text Categorization and Ensemble Pruning in Turkish News Portals. Ph.D. Dissertation. Bilkent Uniiversity, Ankara.
Y. Bengio, P. Simard, and P. Frasconi. 1994. Learning Long-term Dependencies with Gradient Descent is Difficult. Trans. Neur. Netw. 5, 2 (March 1994), 157– 166. https://doi.org/10.1109/72.279181
David M. Blei, Andrew Y. Ng, and Michael I. Jordan. 2003. Latent Dirichlet Allocation. J. Mach. Learn. Res. 3 (March 2003), 993–1022. http://dl.acm.org/ citation.cfm?id=944919.944937
Kyunghyun Cho, Bart van Merrienboer, Çaglar Gülçehre, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation. CoRR abs/1406.1078 (2014). arXiv:1406.1078 http://arxiv.org/abs/1406.1078
Scott Deerwester, Susan T. Dumais, George W. Furnas, Thomas K. Landauer, and Richard Harshman. 1990. Indexing by latent semantic analysis. JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE 41, 6 (1990), 391–407.
M Fatih Amasyali, Aytunç Beken, and Yildiz Teknik Üniversitesi. 2018. Türkçe Kelimelerin Anlamsal Benzerliklerinin Ölçülmesi ve Metin Siniflandirmada Kullanilmasi Measurement of Turkish Word Semantic Similarity and Text Categorization Application. (03 2018).
Aysun Güran, Selim Akyokus, Nilgün Güler Bayazıt, and M Zahid Gürbüz. 2009. (07 2009).
Alex Graves, Abdel-rahman Mohamed, and Geoffrey E. Hinton. 2013. Speech recognition with deep recurrent neural networks. In IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2013, Vancouver, BC, Canada, May 26-31, 2013. 6645–6649. https://doi.org/10.1109/ ICASSP.2013.6638947
Sepp Hochreiter, Yoshua Bengio, Paolo Frasconi, and Jürgen Schmidhuber. 2001. Gradient Flow in Recurrent Nets: the Difficulty of Learning Long- Term Dependencies. (2001).
I.T. Jolliffe. 2002. Principal Component Analysis. Springer.
Daniel Jurafsky and James H. Martin. 2016. Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition (3rd ed.). Prentice Hall PTR, Upper Saddle River, NJ, USA.
Deniz Kılınç, Akın Özçift, Fatma Bozyigit, Pelin Yıldırım, Fatih Yücalar, and Emin Borandag. 2017. TTC-3600: A new benchmark dataset for Turkish text categorization. Journal of Information Science 43, 2 (2017), 174– 185. https://doi.org/10.1177/0165551515620551 arXiv:https://doi.org/10.1177/0165551515620551
Quoc Le and Tomas Mikolov. [n. d.]. Distributed Representations of Sentences and Documents. In In NAACL HLT. 2013.
Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. 2015. Deep learning. Nature 521, 7553 (27 May 2015), 436–444. https://doi.org/10.1038/ nature14539
Yann Lecun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. 1998. Gradient-based learning applied to document recognition. In Proceedings of the IEEE. 2278–2324.
David D. Lewis and Marc Ringuette. 1994. A Comparison of Two Learning Algorithms for Text Categorization. In In Third Annual Symposium on Document Analysis and Information Retrieval. 81–93.
Christopher D. Manning, Prabhakar Raghavan, and Hinrich Schütze. 2008. Introduction to Information Retrieval. Cambridge University Press, New York, NY, USA.
Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 (2013).
Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word representation. In In EMNLP.
G. Salton. 1971. The SMART Retrieval System— Experiments in Automatic Document Processing. Prentice-Hall, Inc., Upper Saddle River, NJ, USA.
Hinrich Schutze, David A. Hull, and Jan O. Pedersen. 1995. A comparison of classifiers and document representations for the routing problem. In ANNUAL ACM CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL - ACM SIGIR. ACM, 229–237.
Jonathan M Cheek Stephen R Briggs. 2007. The role of factor analysis in the development and evaluation of personality scales. (2007).
P. Tüfekci, E. Uzun, and B. Sevinç. 2012. Text classification of web based news articles by using Turkish grammatical features. In 2012 20th Signal Processing and Communications Applications Conference (SIU). 1–4. https://doi.org/10.1109/SIU.2012.6204565
D. Toruno˘glu, E. Çakirman, M. C. Ganiz, S. Akyoku¸s, and M. Z. Gürbüz. 2011. Analysis of preprocessing methods on classification of Turkish texts. In 2011 International Symposium on Innovations in Intelligent Systems and Applications. 112–117. https: //doi.org/10.1109/INISTA.2011.5946084
Filiz Türko˘glu, Banu Diri, and M. Fatih Amasyalı. 2007. Author Attribution of Turkish Texts by Feature Mining. In Advanced Intelligent Computing Theories and Applications. With Aspects of Theoretical and Methodological Issues, De-Shuang Huang, Laurent Heutte, and Marco Loog (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 1086–1093.
Alper Kursat Uysal and Serkan Gunal. 2014. The impact of preprocessing on text classification. Information Processing and Management 50, 1 (2014), 104 – 112. https://doi.org/10.1016/j.ipm.2013.08.006
Sava¸s Yildirim. 2014. A Knowledge-Poor Approach to Turkish Text Categorization. In Computational Linguistics and Intelligent Text Processing, Alexander Gelbukh (Ed.). Springer Berlin Heidelberg, Berlin, Heidelberg, 428–440.
Wenpeng Yin, Katharina Kann, Mo Yu, and Hinrich Schütze. 2017. Comparative Study of CNN and RNN for Natural Language Processing. CoRR abs/1702.01923 (2017). arXiv:1702.01923 http:// arxiv.org/abs/1702.01923