Fast lightweight convolutional neural network for Turkish sentiment analysis

Abstract

This study presents a fast, lightweight, and high-performing fast convolutional neural network (Fast CNN) model tailored for Turkish sentiment analysis (SA). The agglutinative morphology of Turkish, combined with the limited availability of high-quality linguistic resources, introduces significant challenges for conventional approaches. To address these issues, we propose a streamlined Fast CNN architecture consisting of an embedding layer, global max-pooling, dropout, and fully connected layers. Despite its simplicity, the model outperforms seven state-of-the-art convolutional neural network (CNN)-based systems across four benchmark Turkish sentiment datasets. It achieves an average area under the curve (AUC) of 0.94, representing a 6.8% improvement over the strongest baseline and a gain of over 80% relative to several deeper architectures. In addition to its superior accuracy, the model demonstrates reduced computational complexity, making it well-suited for real-world deployment in resourceconstrained environments. Potential applications include customer feedback mining and digital marketing analytics in Turkish-language domains.