Year: 2024
Type: article
Abstract: The rapid proliferation of textual data across online platforms necessitates accurate sentiment analysis. Traditional sentiment analysis methods, which are based on lexical ontology theories and basic rules, have shown limitations in capturing the subtleties and contextual nuances of language. Recent advancements in machine learning and deep learning have shifted the focus toward model-based approaches, yet they often overlook distinct emotional dimensions in varying text structures. To address this issue, we introduce a novel deep neural network architecture that employs multiscale feature extraction and is designed to capture a broad series of emotional features within texts. This approach significantly improves the accuracy of sentiment analysis by effectively discerning subtle emotional nuances. We validate the effectiveness of our proposed model through extensive experiments and comparisons with benchmark methods, demonstrating its superiority in sentiment analysis tasks. Additionally, a detailed ablation study highlights the impact of the multiscale module on the model’s performance.
Source: Innovation & Technology Advances
DOI: 10.61187/ita.v2i2.65
Authors: Hongyu Hu, Jie Zhang, Yang Sun
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4402675486