Article: Campus network public opinion sentiment analysis technology based on XLNet-BiGRU-Att algorithm Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.38 pp.82 - 102 Abstract: To address limitations in campus public opinion sentiment analysis, this study proposes a fusion model combining XLNet, BiGRU, and Attention mechanisms. XLNet captures global semantics, BiGRU models temporal dynamics, and attention focuses on key emotional words, enabling fine-grained analysis. Experiments on 2,500 samples show the model achieves 90.2% accuracy and 89.7% macro-F1, outperforming TextCNN, BERT-BiLSTM, and RoBERTa-GRU by significant margins. In cross-platform transfer (Weibo to Campus Forum), it attains an F1 of 85.4%, demonstrating strong adaptability. With only 20% training data, it maintains 80.1% macro-F1, indicating high data efficiency. Consistency with manual annotation is high, with a Kappa coefficient of 0.957 for negative emotion recognition. The model enhances campus public opinion monitoring and offers insights for complex sentiment analysis tasks. Inderscience Publishers - linking academia, business and industry through research

Title: Campus network public opinion sentiment analysis technology based on XLNet-BiGRU-Att algorithm

Authors: Zheng Qiu

Addresses: School of Marxism, Neijiang Normal University, Neijiang, 641000, China

Abstract: To address limitations in campus public opinion sentiment analysis, this study proposes a fusion model combining XLNet, BiGRU, and Attention mechanisms. XLNet captures global semantics, BiGRU models temporal dynamics, and attention focuses on key emotional words, enabling fine-grained analysis. Experiments on 2,500 samples show the model achieves 90.2% accuracy and 89.7% macro-F1, outperforming TextCNN, BERT-BiLSTM, and RoBERTa-GRU by significant margins. In cross-platform transfer (Weibo to Campus Forum), it attains an F1 of 85.4%, demonstrating strong adaptability. With only 20% training data, it maintains 80.1% macro-F1, indicating high data efficiency. Consistency with manual annotation is high, with a Kappa coefficient of 0.957 for negative emotion recognition. The model enhances campus public opinion monitoring and offers insights for complex sentiment analysis tasks.

Keywords: sentiment analysis; campus public opinion; XLNet; attention mechanism.

DOI: 10.1504/IJICT.2026.153302

International Journal of Information and Communication Technology, 2026 Vol.27 No.38, pp.82 - 102

Received: 13 Oct 2025
Accepted: 13 Jan 2026
Published online: 01 May 2026 *

Title: Campus network public opinion sentiment analysis technology based on XLNet-BiGRU-Att algorithm

Keep up-to-date