Abstract—Opinion mining or Sentiment Analysis (SA) is an essential component of e-commerce applications where consumers generate a large number of reviews. Opinions conveyed about a particular feature of a product have a significant impact on consumer decisions and companies’ reputations. Aspect-based Sentiment Analysis (ABSA), is the process of classifying text according to different aspects and identifying the sentiment associated with each category. In this article, a method is suggested for enhancing the Support Vector Machines (SVM) model to improve its noise tolerance when dealing with the Implicit Aspect Identification (IAI) task which is a subtask of Aspect Based Sentiment Analysis. Using WordNet (WN) semantic relations, modification to the SVM kernel computation is proposed. This study evaluates SVM noise robustness using its classification performance with noisy datasets and multiple kernels. Experiments are conducted on three benchmark datasets of laptops, restaurants, and product reviews. Results are evaluated and analyzed based on the impact of the proposed approach on the performance of SVM for two types of noise (class noise and attribute noise) and two types of kernels (linear kernel and nonlinear kernels). According to the empirical results, the suggested method is shown to increase the noise tolerance of SVM for IAI.
 
Keywords—implicit aspect-based sentiment analysis, support vector machines, wordnet, Lesk algorithm, equalized loss of accuracy, noise robustness, label noise, class noise


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.