Effectiveness of Fine-tuned BERT Model in Classification of Helpful and Unhelpful Online Customer Reviews

The problem of information overload in online review platforms has seriously hampered many customers’ ability to evaluate the quality products or businesses when making purchasing decisions. A large body literature exists that attempts predict helpfulness customer reviews and reported contradictory findings on effectiveness various approaches. Moreover, existing solutions use traditional machine learning techniques handcrafted features, limiting generalization. Therefore, this study aims propose a generalized approach by fine-tuning BERT (Bidirectional Encoder Representations from Transformers) base model. performance BERT-based classifiers is then compared with bag-of-words approaches determine classifiers. evaluations performed using Yelp shopping show fine-tuned outperform classifying helpful unhelpful reviews. In addition, it found sequence length classifier significant impact classification performance.