Bilişim Teknolojileri Dergisi

1307-9697 2147-0715

Gazi Üniversitesi

10.17671/gazibtd.325865

Computer Software Engineering

Bilgisayar Yazılımı Mühendislik

Türkçe Hedef Tabanlı Duygu Analizi İçin Alt Görevlerin İncelenmesi – Hedef Terim, Hedef Kategori Ve Duygu Sınıfı Belirleme

Investigation of Aspect Based Turkish Sentiment Analysis Subtasks – Identification of Aspect Term, Aspect Category And Sentiment Polarity

Çetin

Fatih Samet

İSTANBUL TEKNİK ÜNİVERSİTESİ Eryiğit

Gülşen

İSTANBUL TEKNİK ÜNİVERSİTESİ

01 31 2018

11 1 43 56 07 04 2017 12 13 2017

2008

Bilişim Teknolojileri Dergisi

Gelenekselolarak doküman veya cümle seviyesinde yürütülen duygu analizi çalışmaları,hedef tabanlı duygu analizi çalışmalarının ortaya çıkması ile yeni bir seviyeyetaşınmıştır. Hedef tabanlı duygu analizi (Aspectbased sentiment analysis) kısaca, bir metnin içinde yer alan farklıduyguların ilgili oldukları hedef varlıklar ile birlikte tespit edilmesi olaraktanımlanabilir. Güncel tanımlamalar, hedef tabanlı duygu analizini, üç temelalanla (hedef terim, hedef kategori ve duygu sınıfı) temsil edilen duygutanımlama gruplarını belirlemeyi amaçlayan aşamalı bir görev olarakbetimlemektedir. Bu makalede, Türkçehedef tabanlı duygu analizi konusunda yapılan incelemeler sunulmaktadır. Yürütülençalışmalar, ABSA 2016 yarışmasındaki görevler (1- Hedef kategori belirleme, 2-Hedef terim belirleme, 3- Hedef kategori ve hedef terimin aynı andabelirlenmesi ve 4- Duygu sınıfı belirleme) takip edilerek tasarlanmış ve yineburada sağlanan Türkçe restoran yorumları veri kümesi üzerinde değerlendirilmişlerdir.Hedef kategori, hedef terim ve ikisinin aynı anda belirlenmesi görevleri için,kelime vektörleri (word vectors) ve doğal dil işleme çıktıları (sözcük ve cümleanalizi bilgileri) kullanan koşullu rastgele alanlara (CRF – conditional randomfields) dayalı bir dizilim etiketleme algoritması tasarlanmış ve her üç görevide tek aşamada çözebildiği gösterilmiştir.Elde edilen sonuçlar ile bu ilk üç görev için literatürdeki en yüksekbaşarımların elde edildiği görülmüştür: Hedef kategori belirlemede %66,7F1-skoru, hedef terim belirleme %53,2 F1-skoru, hedef kategori ve hedef teriminaynı anda belirlenmesinde %46,7 F1-skoru. Bunun yanı sıra, duygu sınıfıbelirleme için cümle analizi sonucunda hedef terime yakın kelimelerden özellikseçimine dayalı bir lineer sınıflandırma yöntemi sunulmuş ve literatürde sınırlısistemler tarafından raporlanan en başarılı sonuç (%76,1 F1-skoru) eldeedilmiştir.

Sentiment analysis studies conducted traditionallyat document or sentence level have been moved to a new level with the emergenceof aspect based sentiment analysis studies. Aspect-basedsentiment analysis can be briefly defined as the detection of different opinionscontained within a text together with the target entities to which they relate.Current definitions describe aspect based sentiment analysis as a gradual taskaiming to identify opinion tuples represented by three main fields (targetterm, target category, sentiment class). This article presents ourinvestigations on aspect based Turkish sentiment analysis. The work carried outin this article is designed by following ABSA 2016 competition tasks (1- Aspectcategory identification, 2- Aspect term identification, 3- Identification of aspectcategory and aspect term together and 4- sentiment category classification) andevaluated on the Turkish restaurant reviews dataset provided in the same event.For the first three tasks, a sequence labeling algorithm (based on conditionalrandom fields (CRF)) which uses word vectors and natural language processingoutputs (word and sentence analyses) is proposed and shown to solve these threetasks in one step. Experimental results show that the proposed system achievesthe highest performances for these tasks: 66.7% F1-score for aspect category identification,53.2% F1-score for aspect term identification, 46.7% F1-score for both aspectcategory and aspect term at the same time. Additionally, a linearclassification method based on feature selection from positionally andsyntactically neighboring tokens is proposed for sentiment categoryclassification task and shown to perform as the best constrained systemreported in the literature with 76.1% F1-score.

hedef tabanlı duygu analizi Türkçe doğal dil işleme

aspect based sentiment analysis Turkish language natural language processing

[1] Pang, B. and L. Lee, Opinion mining and sentiment analysis. Foundations and Trends® in Information Retrieval, 2008. 2(1–2): p. 1-135.

[2] Liu, B., Sentiment analysis and opinion mining. Synthesis lectures on human language technologies, 2012. 5(1): p. 1-167.

[3] Pang, B. and L. Lee. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts. in Proceedings of the 42nd Meeting of the Association for Computational Linguistics (ACL'04), Main Volume, Barcelona, Spain. 2004. Association for Computational Linguistics.

[4] Wilson, T., J. Wiebe, and P. Hoffmann. Recognizing contextual polarity in phrase-level sentiment analysis. in Proceedings of the conference on human language technology and empirical methods in natural language processing. 2005. Vancouver, British Columbia, Canada: Association for Computational Linguistics.

[5] Pontiki, M., D. Galanis, H. Papageorgiou, I. Androutsopoulos, S. Manandhar, A.-S. Mohammad, M. Al-Ayyoub, Y. Zhao, B. Qin, O. De Clercq, V. Hoste, M. Apidianaki, X. Tannier, N. Loukachevitch, E. Kotelnikov, N. Bel, S.M. Jimenez-Zafra, and G. Eryiğit, SemEval-2016 Task 5: Aspect Based Sentiment Analysis. Proceedings of the 10th International Workshop on Semantic Evaluation, SemEval ’16, San Diego, California, June 16-17, 2016, 2016: p. 19-30.

[6] Pontiki, M., D. Galanis, H. Papageorgiou, S. Manandhar, and I. Androutsopoulos, SemEval-2015 Task 12: Aspect Based Sentiment Analysis. Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), Denver, Colorado, 2015: p. 486-495.

[7] Pontiki, M., D. Galanis, J. Pavlopoulos, H. Papageorgiou, I. Androutsopoulos, and S. Manandhar, Semeval-2014 task 4: Aspect based sentiment analysis. Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), Dublin, Ireland, 2014: p. 27-35.

[8] Torunoglu, D. and G. Eryigit. A cascaded approach for social media text normalization of Turkish. in Proceedings of the 5th Workshop on Language Analysis for Social Media (LASM)@ EACL. 2014. Gothenburg, Sweden: Association for Computational Linguistics.

[9] Eryiğit, G. and D. Torunoğlu-Selamet, Social Media Text Normalization for Turkish. Natural Language Engineering, 2017(Accepted for publication).

[10] Kaya, M., G. Fidan, and I.H. Toroslu. Sentiment analysis of turkish political news. in Proceedings of the The 2012 IEEE/WIC/ACM International Joint Conferences on Web Intelligence and Intelligent Agent Technology-Volume 01. 2012. Macau, China: IEEE Computer Society.

[11] Yıldırım, E., F.S. Çetin, G. Eryiğit, and T. Temel, The impact of NLP on Turkish sentiment analysis. TÜRKİYE BİLİŞİM VAKFI BİLGİSAYAR BİLİMLERİ ve MÜHENDİSLİĞİ DERGİSİ, 2015. 7(1 (Basılı 8).

[12] Dehkharghani, R., B. YANIKOGLU, Y. SAYGIN, and K. Oflazer, Sentiment analysis in Turkish at different granularity levels. Natural Language Engineering, 2016: p. 1-25.

[13] Eryiğit, G., J. Nivre, and K. Oflazer, Dependency parsing of Turkish. Computational Linguistics, 2008. 34(3): p. 357-389.

[14] Eryigit, G. ITU Turkish NLP Web Service. in EACL. 2014.

[15] Buchholz, S. and E. Marsi. CoNLL-X shared task on multilingual dependency parsing. in Proceedings of the Tenth Conference on Computational Natural Language Learning. 2006. New York City, New York: Association for Computational Linguistics.

[16] Lafferty, J., A. McCallum, and F. Pereira. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. in Proceedings of the eighteenth international conference on machine learning, ICML. 2001. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc.

[17] Alghunaim, A., A Vector Space Approach for Aspect-Based Sentiment Analysis. 2015, Massachusetts Institute of Technology.

[18] Toh, Z. and J. Su, NLANGP at SemEval-2016 Task 5: Improving Aspect Based Sentiment Analysis using Neural Network Features. Proceedings of the 10th International Workshop on Semantic Evaluation, SemEval ’16, San Diego, California, June 16-17, 2016, 2016: p. 282-288.

[19] Okazaki, N., CRFsuite: a fast implementation of conditional random fields (CRFs). 2007.

[20] Şahin, G.G. Turkish Word Embeddings. 2016; Available from: http://isguderg.ml/embedding.html.

[21] Çetin, F.S., E. Yıldırım, C. Özbey, and G. Eryiğit, TGB at SemEval-2016 Task 5: Multi-Lingual Constraint System for As-pect Based Sentiment Analysis. Proceedings of the 10th International Workshop on Semantic Evaluation, SemEval ’16, San Diego, California, June 16-17, 2016, 2016: p. 337-341.

[22] Mohammad, S.M. and P.D. Turney, Nrc emotion lexicon. 2013, NRC Technical Report.

[23] Tamchyna, A. and K. Veselovská, UFAL at SemEval-2016 Task 5: Recurrent Neural Networks for Sentence Classification. Proceedings of the 10th International Workshop on Semantic Evaluation, SemEval ’16, San Diego, California, June 16-17, 2016, 2016: p. 367-371.

[24] Kumar, A., S. Kohail, A. Kumar, A. Ekbal, and C. Biemann, IIT-TUDA at SemEval-2016 Task 5: Beyond Sentiment Lexicon: Combining Domain Dependency and Distributional Semantics Features for Aspect Based Sentiment Analysis. Proceedings of the 10th International Workshop on Semantic Evaluation, SemEval ’16, San Diego, California, June 16-17, 2016, 2016: p. 1129-1135.

[25] Ruder, S., P. Ghaffari, and J.G. Breslin, INSIGHT-1 at SemEval-2016 Task 5: Deep Learning for Multilingual Aspect-based Sentiment Analysis. Proceedings of the 10th International Workshop on Semantic Evaluation, SemEval ’16, San Diego, California, June 16-17, 2016, 2016.

[26] Dehkharghani, R., Y. Saygin, B. Yanikoglu, and K. Oflazer, SentiTurkNet: a Turkish polarity lexicon for sentiment analysis. Language Resources and Evaluation, 2016. 50(3): p. 667-685.

[27] Mohammad, S. NRC Word-Emotion Association Lexicon. 2013; Available from: http://saifmohammad.com/WebPages/NRC-Emotion-Lexicon.htm.

[28] Şeker, G.A. and G. Eryiğit, Extending a CRF-based named entity recognition model for Turkish well formed text and user generated content. Semantic Web Journal, 2017(doi:10.3233/SW-170253).