Scholarlink

<p>Cross-domain aspect classification aims to determine and classify aspects of a product over different domains or categories. To attain insights into user opinions and experiences across several domains, the objective is to categorize and analyze the various features and attitudes connected with them such as customer reviews and opinions. Despite that, the existing works are limited by several challenges related to low classification accuracy, feature generalization, computational inefficiency, and label sparsity. To overcome these problems, this research work develops a novel DL methodology named Knowledge Informed Cascaded Attention Transformer (KICAT) for an effective cross-domain aspect classification process. The proposed KICAT technique develops a Knowledge-based Aspect-aware sentence generation (K-ASG) model by integrating knowledge injection and aspect-aware sentence generation procedures for generating aspect-aware sentences and masked sentences. Furthermore, the cascaded attention transformer technique is developed to capture the contextual features such as sentences, words, and texts from the K-ASG model and introduces cascaded group attention for reducing computational complexity. The extracted features are classified into multiple cross-domain contextualized aspects using a multi-layer perceptron classifier. The experimental findings show the KICAT framework precisely classifies the aspects into different polarities such as positive, negative, and neutral. The KICAT technique attains remarkable outcomes including an accuracy of 98.89% and low execution time of 2.5&#xa0;s. From this experimental validation, it’s proven that the KICAT methodology obtains excellent cross-domain aspect categorization results and enables businesses and scholars to learn more insights about customer opinions and viewpoints.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

KICAT: achieving high accuracy in cross-domain aspect classification with a novel transformer approach

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • T. Kumaragurubaran,
  • S. Vinodh Kumar,
  • P. Kumar,
  • S. Dhanasekaran

摘要

Cross-domain aspect classification aims to determine and classify aspects of a product over different domains or categories. To attain insights into user opinions and experiences across several domains, the objective is to categorize and analyze the various features and attitudes connected with them such as customer reviews and opinions. Despite that, the existing works are limited by several challenges related to low classification accuracy, feature generalization, computational inefficiency, and label sparsity. To overcome these problems, this research work develops a novel DL methodology named Knowledge Informed Cascaded Attention Transformer (KICAT) for an effective cross-domain aspect classification process. The proposed KICAT technique develops a Knowledge-based Aspect-aware sentence generation (K-ASG) model by integrating knowledge injection and aspect-aware sentence generation procedures for generating aspect-aware sentences and masked sentences. Furthermore, the cascaded attention transformer technique is developed to capture the contextual features such as sentences, words, and texts from the K-ASG model and introduces cascaded group attention for reducing computational complexity. The extracted features are classified into multiple cross-domain contextualized aspects using a multi-layer perceptron classifier. The experimental findings show the KICAT framework precisely classifies the aspects into different polarities such as positive, negative, and neutral. The KICAT technique attains remarkable outcomes including an accuracy of 98.89% and low execution time of 2.5 s. From this experimental validation, it’s proven that the KICAT methodology obtains excellent cross-domain aspect categorization results and enables businesses and scholars to learn more insights about customer opinions and viewpoints.