Sentiment classification for Turkish Twitter feeds using LDA

Abstract

Today, people are able to express their opinions and ideas on any subject using personal blogs or a variety of social media environments. Contents shared by users with each passing day increases the data size on the web but this situation is a complication for meaningful information extraction. Therefore, sentiment analysis studies aimed at automatically obtaining meaningful information from social media content showed a significant increase in recent years. In these studies, automatically analyzing the content of messages and identifying sentiment such as positive, neutral, negative etc. are intended. In this study, a system has been proposed for analyzing sentiment of Turkish Twitter feeds. In addition, sentiment classification which is a leading problem for sentiment analysis studies has been performed using topic modeling and the effect of modeling on results are examined. The topic model has been created using LDA (Latent Dirichlet Allocation) algorithm. Experimental results obtained with different feature extraction model and classifiers. The results show that topic modeling is more successful compared to the methods used in the previous studies. Also, sentiment classification based on topic model knowledge has been found to increase sentiment analysis success by %26 compared to our previous work.