Scholarlink

<p>Optimization is at the heart of machine learning refinement in the age of big data and AI. Although natural language processing (NLP) has achieved remarkable progress, sentiment analysis in semi-supervised settings still shows a significant challenge in having limited labeled data. Semi-supervised learning can be used to overcome this limitation as it utilizes both labeled and unlabeled data, yet the success of the semi-supervised model is highly dependent on feature selection that reduces noise, increases interpretability, and improves classification accuracy. Classical feature selection methods struggle with providing a finer clarity as well as redundancy detection, it requires implementing supervised feature refinement approaches for better sentiment classification. To resolve these issues, the proposed work introduces a semi-supervised e-commerce review sentiment analysis framework utilizing Neutrosophic Fuzzy Chi-Square (NFCS) for uncertainty-aware feature selection, as well as the Firefly Algorithm for feature optimization. For text-based opinions, the experimental progress of CNN architectures such as AlexNet and VGGNet (designed for image processing tasks) that use text-based sentiment classification and hierarchical feature representations to learn text-based rules that summarize and localize textual dependencies. This framework consists of self-training and co-training-based methods that improve the process. A conditional probability-based validation mechanism is further used to prevent misclassification during the pseudo-labeling process, so as to verify reliable knowledge transfer from unlabeled data. Extensive experimental comparisons establish the combined use of deep learning-based feature extraction, Neutrosophic Fuzzy feature selection, and optimization using Firefly Algorithm have a significant impact on sentiment classification. The main novelty of this work results from the incorporation of uncertainty-aware feature selection in a metaheuristic optimization framework integrated with deep learning within a semi-supervised scenario that leads to both a scalable and interpretable sentiment analytical approach with low computational cost for e-commerce implementations.</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

Feature selection for semi-supervised sentiment analysis of e-commerce reviews using CNN and neutrosophic fuzzy parameters

  • 期刊论文

  • 发表日期：

    2025年4月24日

查看全文
  • Alok Kumar Jena,
  • K. Murali Gopal,
  • Abinash Tripathy,
  • Siba Prasada Tripathy

摘要

Optimization is at the heart of machine learning refinement in the age of big data and AI. Although natural language processing (NLP) has achieved remarkable progress, sentiment analysis in semi-supervised settings still shows a significant challenge in having limited labeled data. Semi-supervised learning can be used to overcome this limitation as it utilizes both labeled and unlabeled data, yet the success of the semi-supervised model is highly dependent on feature selection that reduces noise, increases interpretability, and improves classification accuracy. Classical feature selection methods struggle with providing a finer clarity as well as redundancy detection, it requires implementing supervised feature refinement approaches for better sentiment classification. To resolve these issues, the proposed work introduces a semi-supervised e-commerce review sentiment analysis framework utilizing Neutrosophic Fuzzy Chi-Square (NFCS) for uncertainty-aware feature selection, as well as the Firefly Algorithm for feature optimization. For text-based opinions, the experimental progress of CNN architectures such as AlexNet and VGGNet (designed for image processing tasks) that use text-based sentiment classification and hierarchical feature representations to learn text-based rules that summarize and localize textual dependencies. This framework consists of self-training and co-training-based methods that improve the process. A conditional probability-based validation mechanism is further used to prevent misclassification during the pseudo-labeling process, so as to verify reliable knowledge transfer from unlabeled data. Extensive experimental comparisons establish the combined use of deep learning-based feature extraction, Neutrosophic Fuzzy feature selection, and optimization using Firefly Algorithm have a significant impact on sentiment classification. The main novelty of this work results from the incorporation of uncertainty-aware feature selection in a metaheuristic optimization framework integrated with deep learning within a semi-supervised scenario that leads to both a scalable and interpretable sentiment analytical approach with low computational cost for e-commerce implementations.