A hybrid deep learning model for analyzing the sentiments of products

Purpose
The rapid growth of web-based applications, especially digital networking sites and E-commerce platforms, has led to an influx of user reviews, prompting the need for sentiment analysis. Aspect-based sentiment analysis (ABSA) helps identify sentiment tendencies toward specific aspects of products or services, though challenges like noisy, informal reviews and limitations in traditional feature extraction methods persist.


Design/methodology/approach
The model integrates the Transformer-based DeBERTa and deep learning-based IDCNN for effective aspect-level feature extraction from review data. Sentiment classification is performed using an attention-based BiLSTM-CRF model, combining bidirectional long short-term memory (BiLSTM) to capture contextual dependencies with a conditional random field (CRF) layer for refining output.


Findings
Experimental results across four benchmark datasets demonstrate that the proposed hybrid model consistently outperforms existing approaches. The model achieved accuracy scores of 93.08% on DS-I, 90.21% on DS-II, 88.76% on DS-III, and 92.86% on DS-IV, indicating its strong performance in aspect-based sentiment analysis, particularly in handling noisy user reviews.


Originality/value
This work introduces a novel approach by combining DeBERTa and IDCNN for improved aspect-level feature extraction and enhancing sentiment classification with an attention-based BiLSTM-CRF model. This innovation provides a more effective solution for sentiment analysis in the context of user-generated content.