Scholarlink

<p>The availability of the Internet and various social media platforms allows users to share their opinions on any products or issues. These comments serve as important resource for understanding customers’ feelings about a product. Sentiment analysis plays a crucial role in collecting, processing, and interpreting this information. While conducting the sentiment analysis on reviews, many researchers prefer binary classification; however, this approach can oversimplify the analysis. Therefore, this study proposes a multi-class sentiment analysis with five distinct output categories. Two different datasets, namely IMDb and the Polarity dataset, are utilized for the analysis. The IMDb dataset provides a separate training and testing dataset, while the Polarity dataset does not have such a separation. Thus, 10-fold cross-validation approach is adopted for the Polarity dataset sentiment analysis. After selecting the datasets, the reviews undergo a pre-processing phase and are transformed into a matrix format for machine learning analysis. The reviews represented in the matrix format are then processed using Particle Swarm Optimization (PSO) and Flower Pollination Algorithm (FPA) for feature selection. The selected features are further processed using an Artificial Neural Network (ANN) with variable hidden nodes to achieve binary classification results. Finally, these obtained binary classification results are processed through a Fuzzy rule-based system to obtain five-class analysis result. The accuracy achieved by combining PSO with ANN is 98.5% for the IMDb dataset and 98.7% for the Polarity dataset. Meanwhile, the combination of FPA and ANN yields an accuracy of 98.2% for the IMDb dataset and 98.4% for the Polarity dataset.</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

Multi-class sentiment analysis using hybrid nature inspired machine learning technique and fuzzy rule-based system

  • 期刊论文

  • 发表日期：

    2026年5月16日

查看全文
  • Abinash Tripathy

摘要

The availability of the Internet and various social media platforms allows users to share their opinions on any products or issues. These comments serve as important resource for understanding customers’ feelings about a product. Sentiment analysis plays a crucial role in collecting, processing, and interpreting this information. While conducting the sentiment analysis on reviews, many researchers prefer binary classification; however, this approach can oversimplify the analysis. Therefore, this study proposes a multi-class sentiment analysis with five distinct output categories. Two different datasets, namely IMDb and the Polarity dataset, are utilized for the analysis. The IMDb dataset provides a separate training and testing dataset, while the Polarity dataset does not have such a separation. Thus, 10-fold cross-validation approach is adopted for the Polarity dataset sentiment analysis. After selecting the datasets, the reviews undergo a pre-processing phase and are transformed into a matrix format for machine learning analysis. The reviews represented in the matrix format are then processed using Particle Swarm Optimization (PSO) and Flower Pollination Algorithm (FPA) for feature selection. The selected features are further processed using an Artificial Neural Network (ANN) with variable hidden nodes to achieve binary classification results. Finally, these obtained binary classification results are processed through a Fuzzy rule-based system to obtain five-class analysis result. The accuracy achieved by combining PSO with ANN is 98.5% for the IMDb dataset and 98.7% for the Polarity dataset. Meanwhile, the combination of FPA and ANN yields an accuracy of 98.2% for the IMDb dataset and 98.4% for the Polarity dataset.