Sentiment Analysis of Online Customer Reviews Using Modern Natural Language Processing Approaches

Abstract

In this study, sentiment analysis was conducted using Turkish online customer reviews of four seafood restaurants based in İzmir. Modern natural language processing approaches were employed as part of the analysis, including a Turkish sentiment analysis model for BERT and multilingual models within the framework of zero-shot text classification. In addition, large language models (LLMs) such as OpenAI 4o, Gemini 2.0 Flash, and DeepSeek V3 were evaluated. Model performance was assessed using evaluation metrics, including accuracy, precision, recall, and F1 score. The findings indicate that LLMs—particularly DeepSeek V3—demonstrated high performance and could effectively process contextual representations even in unlabelled datasets. Furthermore, sentiment trends towards restaurants over a four years were analysed to track temporal changes in customer satisfaction. This approach revealed temporal performance differences among the restaurants over time and enabled the development of sustainable improvement strategies aligned with customer expectations. The proposed method offers a fast and data-driven solution to assist managers in monitoring customer satisfaction, evaluating service quality, and identifying underlying causes of dissatisfaction, supporting strategic decision-making processes and contributing to corporate image management.