Sentiment Polarity Detection in Azerbaijani Social News Articles

Proceedings of Recent Advances in Natural Language Processing, pages 703–710, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_082 703 Sentiment Polarity Detection in Azerbaijani Social News Articles Sevda Mammadli1, Shamsaddin Huseynov1, Huseyn Alkaramov1, Ulviyya Jafarli1, Umid Suleymanov2, Samir Rustamov1,3 1 School of Information Technologies and Engineering, ADA University, Baku, Azerbaijan 2 E-GOV Development Center, Baku, Azerbaijan, 3 Institute of Control Systems, Azerbaijan {smammadli2019, shuseynov2019, halkaromov2019, ujafarli2019}@ada.edu.az umidsuleymanov96@gmail.com srustamov@ada.edu.az Abstract Text classification field of natural lan- guage processing has been experiencing remarkable growth in recent years. Espe- cially, sentiment analysis has received a considerable attention from both industry and research community. However, only a few research examples exist for Azerbaija- ni language. The main objective of this re- search is to apply various machine learn- ing algorithms for determining the senti- ment of news articles in Azerbaijani lan- guage. Approximately, 30.000 social news articles have been collected from online news sites and labeled manually as nega- tive or positive according to their senti- ment categories. Initially, text prepro- cessing was implemented to data in order to eliminate the noise. Secondly, to convert text to a more machine-readable form, BOW (bag of words) model has been ap- plied. More specifically, two methodolo- gies of BOW model, which are tf-idf and frequency-based model have been used as vectorization methods. Additionally, SVM, Random Forest, and Naive Bayes algo- rithms have been applied as the classifica- tion algorithms, and their combinations with two vectorization approaches have been tested and analyzed. Experimental re- sults indicate that SVM outperforms other classification algorithms. 1 Introduction Development of technology generates a vast amount of data flow through the internet and en- courages the creation of sophisticated methodolo- gies to store and analyze it. Analyzing such huge volumes of data manually could lead to the waste of time and investment. Therefore, currently more automated and efficient ways are implemented to solve the problem. With the growth of the produced data, a branch of Artificial Intelligence - Natural Language Pro- cessing (NLP) had begun to evolve. Text classifi- cation is a fundamental part of NLP and has been applied in many areas. The aim of text classifica- tion is to group data into predefined categories based on the labeled data using various machine learning techniques. It requires several stages to categorize the data including data collection, pre- processing, and feature extraction. One of the im- plementation areas of text classification includes sentiment analysis. Sentiment analysis helps to define whether an author’s opinion towards a spe- cific topic is negative or positive. Considering the fact that people’s opinion directly influences the businesses and organizations it is no surprise that sentiment analysis receives a lot of attention. Although sentiment analysis has been applied largely worldwide, research on its usage and utili- ty on Azerbaijani language is scarce. English lan- guage has the luxury of having numerous annotat- ed datasets as well as having well-tuned text pre- processing techniques. Different from English language, natural language processing algorithms are not improved sufficiently in Azerbaijani which is an …

Content truncated to 500 words