Breaking down linguistic complexities: A structured approach to aspect-based sentiment analysis
Kanwal Ahmed,Muhammad Imran Nadeem,Zhiyun Zheng,Dun Li,Inam Ullah,Muhammad Assam,Yazeed Yasin Ghadi,Heba G. Mohamed
DOI: https://doi.org/10.1016/j.jksuci.2023.101651
IF: 9.006
2023-07-26
Journal of King Saud University - Computer and Information Sciences
Abstract:Aspect-based sentiment analysis refers to the task of determining the sentiment polarity associated with particular aspects mentioned in a sentence or document. Previous studies have used attention-based neural network models to connect aspect terms with context words, but these models often perform poorly due to limited interaction between aspect terms and opinion words. Furthermore, these models typically focus only on explicitly stated aspect objects, which can be overly restrictive in certain scenarios. Current sentiment analysis methods that rely on aspect categories also often fail to consider the implicit placement of aspect-category information within the context. While existing models may produce strong results, they often lack domain knowledge. To address these issues, this study proposes an Aspect-position and Entity-oriented Knowledge Convolutional Graph (APEKCG) consisting of two modules: the Aspect position-aware module (APA) and the Entity oriented Knowledge Dependency Convolutional Graph (EKDCG). The APA module is designed to integrate aspect-specific sentiment features for sentiment classification by incorporating information about aspect categories into different parts of the context. The EKDCG module incorporates entity-oriented knowledge, dependency labels, and syntactic path using a dependence graph. Experimental results on five benchmarks Natural Language Processing (NLP) datasets of the English language demonstrate the effectiveness of the proposed APEKCG framework. Furthermore, the APEKCG outperformed previous state-of-the-art models with its accuracy, achieving 89.13%, 84.32%, 89.02%, 79.64%, and 90.22% on the MAMS, Laptop, Restaurant, AWARE, and SemEval-15&16 datasets, respectively.
computer science, information systems