Classification by Feature Fusion: The Impact of Linguistic Features on Sentiment Analysis in Turkish Texts

The rapid rise of e-commerce platforms has transformed user reviews into a significant data source. While these reviews present an important opportunity to understand user sentiments, conducting sentiment analysis in the Turkish language poses various challenges. The complex linguistic structure of Turkish and the limited number of studies in the literature restrict the accuracy and scope of analyses in this domain. Particularly, punctuation marks, grammatical structures, and other linguistic features are key aspects to consider during the sentiment analysis process. In this study, the TRSAv1 dataset was utilized to examine the impact of linguistic features specific to the Turkish language on sentiment classification. Morphological, syntactic, and semantic features were extracted from the texts, and their contributions to classification performance were evaluated. The study analyzed the influence of linguistic differences across various sentiment classes. A classification accuracy of 89% was achieved for positive-negative sentiment analysis, whereas distinguishing between positive and neutral sentiments was found to be more challenging, resulting in a lower performance. This study aims to contribute to the literature by revealing the effects of Turkish-specific linguistic features on sentiment analysis, thereby addressing a critical gap in this area of research.