Hierarchical aspect-based sentiment analysis using semantic capsuled multi-granular networks

Abstract

In the ever-evolving domain of sentiment analysis, discerning intricate sentiments towards specific aspects and their sub-components within textual data has become pivotal. This paper introduces the Semantic Capsuled Hierarchical Multi-Granular Network (SCH-MGN) model, an innovative approach explicitly designed for aspect-based sentiment analysis (ABSA) challenges. The SCH-MGN model is primed to evaluate sentiments at both macro (broader topics) and micro (detailed sub-aspects) hierarchical levels, offering a comprehensive sentiment evaluation spectrum. By integrating mechanisms like the Semantic Knowledge Graph Attention Network (SKG-AN) for targeted aspect extraction, Hierarchical Embedding Layers leveraging Multilingual BERT (mBERT), and advanced neural architectures including Recurrent Neural Networks (RNNs) and Temporal Convolutional Networks (TCNs), the model ensures a nuanced sentiment interpretation. The paper provides a meticulous dissection of the model's methodology, from tokenization and embedding to detailed sentiment extraction, accentuating its capability to offer granular sentiment interpretations. Empirical illustrations validate the model's proficiency in handling compound sentiments, cementing its potential as an indispensable tool for businesses, reviewers, and analysts. This groundbreaking approach to ABSA promises to redefine the granularity with which we understand and evaluate textual sentiments in diverse domains.