RESEARCH ON THE USE OF MODELING TECHNIQUES FOR ANALYZING CUSTOMER FEEDBACK

Abstract

The exponential growth of social media platforms and applications on the Internet has resulted in a staggering amount of user-generated textual content, including comments and reviews. Consequently, users often face difficulties in drawing valuable conclusions or relevant information from such content. To solve this problem, machine learning and natural language processing algorithms have been applied to analyze the vast amount of textual data available on the Internet. In recent years, topic modeling techniques have gained considerable popularity in this field. In this study, we comprehensively investigate and compare five commonly used topic modeling techniques that are specifically applied to customer reviews. The methods studied are: Latent Semantic Analysis (LSA), Latent Dirichlet Allocation (LDA), Nonnegative Matrix Factorization (NMF), Pachinko Allocation Model (PAM), Top2Vec, and BERTopic. By practically demonstrating their advantages in identifying important topics, we aim to emphasize their effectiveness in real-world scenarios. To evaluate the effectiveness of these topic modeling techniques, we carefully select two text datasets. The evaluation is based on standard statistical evaluation metrics such as the topic consistency score. Our findings show that BERTopic consistently produces more meaningful topic extractions and achieves favorable results.