Enhancing Restaurant Customer Review Analysis: Multi-Class Text Classification with BERT

Online customer reviews can be valuable information when processed with the right techniques. Businesses aim to enhance their strategies in managing operations by leveraging sentiment analysis from their clientele. In a business context, the primary focus of SA is to capture individual perspectives on products. Presently, sentiment analysis not only deals with emotional polarity but also classifies sentiments towards various product-related aspects. This research aims to classify restaurant customer review data on Google Maps into categories such as “menu,” “taste,” “indoor atmosphere,” and “outdoor atmosphere”. We utilize 450 entries for this dataset. The model used is Bidirectional Encoder Representation from Transformers (BERT), a technique based on transformer-based neural networks, capable of learning grammatical structures, interpreting semantics, and deriving relationships among contextual entities. The results show a less satisfying accuracy score of 41.11
⁰
/0. This is due to several inaccuracies in the label annotations of the utilized data. Further analysis is discussed in the Discussion and Conclusion section.