CogniRetail: A Contextual Semantic Inference Architecture for MultiResolution Consumer Opinion Intelligence

Abstract

The rapid expansion of e-commerce platforms has resulted in an unprecedented increase in online product reviews, making sentiment analysis an essential tool for understanding customer opinions and supporting business decision-making. These reviews contain valuable information regarding customer experiences, product quality, and purchasing preferences. However, manually analyzing large volumes of textual feedback is timeconsuming, inconsistent, and incapable of handling continuously growing datasets. In addition, conventional sentiment classification methods often struggle to capture contextual semantics and perform effectively when sentiment classes are unevenly distributed. To address these challenges, this research proposes an intelligent sentiment classification framework that integrates advanced semantic representation with hybrid deep learning techniques. Initially, customer reviews undergo comprehensive preprocessing and Exploratory Data Analysis (EDA) to improve data quality and examine the underlying characteristics of the dataset. Subsequently, Sentence Bidirectional Encoder Representations from Transformers (SBERT) generates contextual embeddings that preserve semantic relationships within the review text. To alleviate class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is employed to generate representative samples for minority sentiment categories, thereby improving model generalization. The extracted features are processed using a Deep Neural Network (DNN)-based feature selection mechanism combined with a Boosted Rules Classifier (BRC) to classify customer opinions into Negative, Neutral, and Positive sentiment classes. For performance comparison, the proposed framework is evaluated against Random Forest Classifier (RFC), Light Gradient Boosting Machine (LGBM), and Extreme Gradient Boosting (XGBoost). Experimental results demonstrate that the proposed model consistently achieves higher classification accuracy and reduced prediction bias, providing a scalable and reliable solution for customer sentiment analysis, product improvement, marketing optimization, and customer relationship management.