Article: New media public opinion classification technology integrating improved feature selection algorithm and multi-channel random undersampling Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.78 pp.1 - 29 Abstract: A new media public opinion classification model is proposed to address limitations of traditional methods, including high-dimensional sparseness, poor semantic understanding, and category imbalance. The model integrates a three-branch decision feature selection algorithm (combining mutual information and improved term-category information) with a multi-channel random undersampling mechanism. A hybrid architecture combining Transformer, LSTM, and CNN captures global context, sequence dependence, and local features. Experiments on Weibo and Douyin datasets achieve accuracies of 89.47% and 87.93%, with macro F1 values of 83.26% and 81.17%, respectively. The model improves negative opinion recognition F1 by 6.97% over a pre-trained language model, with inference time of 28.5ms (120 samples) and stable 15.3% CPU usage. This work provides an accurate, efficient, and robust engineering solution for new media public opinion monitoring. Inderscience Publishers - linking academia, business and industry through research

Title: New media public opinion classification technology integrating improved feature selection algorithm and multi-channel random undersampling

Authors: Jinming Zheng; Chanjuan Lin; Xiao Yu; Tingjiao Ye; Yanyan Li; Chen Zhan

Addresses: Institute of Digital and Intelligent Technology, Wenzhou Vocational College of Science and Technology (Wenzhou Academy of Agricultural Sciences), Wenzhou, 325000, China; School of Multimedia Technology and Communication, University Utara Malaysia, Sintok, 06010, Malaysia ' Institute of Digital and Intelligent Technology, Wenzhou Vocational College of Science and Technology (Wenzhou Academy of Agricultural Sciences), Wenzhou, 325000, China ' Institute of Digital and Intelligent Technology, Wenzhou Vocational College of Science and Technology (Wenzhou Academy of Agricultural Sciences), Wenzhou, 325000, China ' Development and Planning Department, Wenzhou Vocational College of Science and Technology (Wenzhou Academy of Agricultural Sciences), Wenzhou, 325000, China ' Development and Planning Department, Wenzhou Polytechnic, Wenzhou, 325000, China ' Institute of Digital and Intelligent Technology, Wenzhou Vocational College of Science and Technology (Wenzhou Academy of Agricultural Sciences), Wenzhou, 325000, China

Abstract: A new media public opinion classification model is proposed to address limitations of traditional methods, including high-dimensional sparseness, poor semantic understanding, and category imbalance. The model integrates a three-branch decision feature selection algorithm (combining mutual information and improved term-category information) with a multi-channel random undersampling mechanism. A hybrid architecture combining Transformer, LSTM, and CNN captures global context, sequence dependence, and local features. Experiments on Weibo and Douyin datasets achieve accuracies of 89.47% and 87.93%, with macro F1 values of 83.26% and 81.17%, respectively. The model improves negative opinion recognition F1 by 6.97% over a pre-trained language model, with inference time of 28.5ms (120 samples) and stable 15.3% CPU usage. This work provides an accurate, efficient, and robust engineering solution for new media public opinion monitoring.

Keywords: new media public opinion classification; three-way decision feature selection; multi-channel random undersampling; imbalanced data classification.

DOI: 10.1504/IJICT.2026.155024

International Journal of Information and Communication Technology, 2026 Vol.27 No.78, pp.1 - 29

Received: 17 Dec 2025
Accepted: 16 Mar 2026
Published online: 22 Jul 2026 *

Title: New media public opinion classification technology integrating improved feature selection algorithm and multi-channel random undersampling

Keep up-to-date