A multi-label text classifier approach for understanding electronic word-of-mouth of restaurants on Google Maps

Abstract

Restaurant owners need to understand and react to customer feedback to remain competitive in the long term. Customers provide essential feedback electronically via online platforms such as Google Maps. To better understand customer feedback, we developed a multi-label text classifier to classify feedback into categories of aspects customers criticize and comment on. Since restaurants, like many small and medium-sized enterprises, do not have the resources to maintain computationally intensive deep learning architectures, we present a simple knowledge distillation approach in this paper. On the test dataset, our approach performs better than a BERT model at a much smaller model size and with significantly better inference time. These results provide a novel approach to understanding electronic word of mouth for small and medium-sized enterprises.