Classification of Ecommerce Customer Reviews Using DistilBERT with Support Vector Machine

In the rapid evolving market, the reviews of product and services can influence the purchasing decision of a customer. Therefore, manually evaluating the reviews is the most challenging tasks to make the decision and observe the business model. The Machine Learning (ML) and Deep Learning (DL) model are introduced for the customer reviews classification. However, the existing models such as Bidirectional Encoder Representations from Transformers (BERT) suffered with computational complexity and slow interference. In this paper, distilBERT with Support Vector Machine (distilBERT-SVM) is proposed for the effective classification of ecommerce customer reviews. Initially, women's ecommerce clothing reviews dataset is considered and then preprocessed with punctuation, stop word removal and lemmatization. Then, dB model is introduced for extracting the features using lexical and transformer encoders for converting the vectors into word, segment and a positional embedding's. After that, linear layer is served as a feature extraction layer and these features are further processed into classification with SVM classifier to classify the reviews into binary labels as positive and negative reviews. From the results, the proposed distilBERT-SVM achieved better results than the existing systems in terms of accuracy, precision, recall and Fl-score of 98.2%, 98.5%, 92.3% and 94.5% respectively.