HDR‐SA: A Hybrid Deep Learning and RoBERTa‐Based Framework for Sentiment and Aspect Analysis

The ability to comprehend complex viewpoints in text is critical for sentiment analysis (SA), particularly at the aspect level, yet existing models struggle with accurately identifying sentiment polarities and aspect‐specific expressions due to their reliance on large, manually annotated, domain‐specific datasets. To address these challenges, this paper introduces hybrid deep learning and RoBERTa‐based SA (HDR‐SA), a novel hybrid deep learning framework that integrates convolutional neural networks (CNNs), bidirectional long short‐term memory (BiLSTM) networks, and the RoBERTa transformer model to perform comprehensive sentiment and aspect analysis. The proposed model begins with rigorous data preprocessing and normalization, utilizes Valence Aware Dictionary and sEntiment Reasoner (VADER) for sentiment scoring, constructs embedding vectors via Word2Vec, and employs a CNN‐BiLSTM architecture enhanced by RoBERTa to capture both sequential and contextual embeddings for refined sentiment classification. The novelty of HDR‐SA lies in its hybrid integration of conventional natural language processing (NLP) techniques with deep learning and transformer‐based contextual understanding, enabling robust SA without the extensive need for domain‐specific annotated data. Evaluated on the large‐scale 515K Hotel Reviews dataset, HDR‐SA achieved an accuracy of 95.75%, a precision of 0.96, a recall of 0.97, and an F1‐score of 0.96, outperforming contemporary models such as target‐dependent LSTM (TD‐LSTM), ResNet‐SCSO, and CNN‐GA. These results demonstrate HDR‐SA’s effectiveness in aspect‐level SA and its scalability across diverse domains while reducing dependency on annotated resources.