Multi-Class Sentiment Analysis from Turkish Tweets with RNN

Twitter is a social media platform where users can post their messages called ‘tweets’. Comment on a product, person, or event on Twitter; It takes reading and interpreting thousands of tweets to find out what emotion it represents. With sentiment analysis, it is possible to perform this process automatically in a short time. In this study; A data set consisting of Turkish tweets divided into 5 different emotion categories was used. Sentiment analysis was carried out using RNN architecture, which is a deep learning method. In the dataset, there are equal numbers of tweets for each of the emotions “angry”, “fear”, “happy”, “surprise”, “sad”. The success of the models established by performing multi-class sentiment analysis with LSTM, BiLSTM and GRU based on RNN architecture was compared. Highest accuracy; It has been in the model established with bidirectional LSTM, that is, BiLSTM, which is very successful in past and future word contexts.