Words, Meanings, Characters in Sentiment Analysis

In the sentiment analysis, text representation is the most important research topic as another natural language processing tasks. In this study, word, semantic and character based representations of texts are compared for Turkish sentiment analysis. Classical and deep learning based approaches are used in data modeling. Character-based representations are more successful with both classical and deep learning based approaches. Moreover, a large-scale tagged data set has been presented for Turkish sentiment analysis researchers.