Revealing Hidden Sentiments: Implicit Aspectbased Sentiment Analysis Using Bert on Amazon Reviews

Traditional Sentiment Analysis (SA) detects sentiment in text but often overlooks the aspects that drive it. Aspect-Based Sentiment Analysis (ABSA) links sentiments to aspects, yet many are implicit and require contextual inference, something conventional models often miss. This study introduces a Bidirectional Encoder Representations from Transformers (BERT) model for joint implicit aspect extraction and sentiment classification, enabling the detection of hidden aspects and their associated sentiments. Its novelty is leveraging contextual inference to identify hidden aspect-sentiment pairs, which conventional ABSA models often fail to capture. The model is tested on Amazon reviews across four product categories: personal care, home care, foods, and refreshments. It achieves 86 % accuracy, outperforming the baseline models. Results show “Scent” dominates in personal and home care, while “Flavour/Taste” is most frequent in food and refreshments. Aspects like “Price” and “Product Effectiveness” show mixed sentiment trends. By uncovering context-dependent cues, the model improves sentiment analysis depth and reliability. Its practical value lies in helping businesses understand what consumers feel and why, even when opinions are implicit. This enables better product design, targeted marketing, and user experience across e-commerce and social media domains.