A COMPARATIVE STUDY OF SENTIMENT CLASSIFICATION: TRADITIONAL NLP VS. NEURAL NETWORK APPROACHES

Abstract

The current research compares traditional natural language processing methods, such as Naive Bayes and Support Vector Machine, to neural network approaches, particularly Multi-Layer Perceptron, to classify positive and negative sentiments regarding company customer service. This research is motivated by the need to understand the effectiveness of these two approaches in analyzing and classifying sentiment in customer reviews, a crucial aspect of enhancing the quality of customer service. The author evaluated accuracy, speed, and adaptability to complex and diverse review content using a dataset containing various business customer reviews. The findings of this study indicate that neural network approaches, particularly Multi-Layer Perceptron, tend to provide superior performance in classifying customer sentiment with greater precision, albeit at a higher computational cost. Traditional methods such as Naive Bayes and Support Vector Machine still apply in situations with limited resources. The results of this research provide valuable guidance for companies in selecting an appropriate approach to analyzing customer sentiment, with the potential to increase understanding of customer views and improve overall customer service. Nave Bayes achieves 68.75% accuracy, Support Vector Machine achieves 87.5% accuracy, and Multi-Layer Perceptron achieves 100% accuracy.