To address the complex nonlinear characteristics of information dissemination in Weibo social hot topics, as well as the shortcomings of traditional monitoring methods—such as insufficient semantic understanding, experience-dependent feature engineering, and lagging warnings—this paper designs and implements a learning-based intelligent analysis and early warning system. First, a deep semantic representation model for Weibo text is constructed based on BERT to handle issues of short, noisy, and contextually complex text. Second, a hotspot identification and sentiment evolution analysis method is developed by integrating HDBSCAN density clustering with BiLSTM-Attention. Third, a quantitative hotspot calculation model combining user features, content features, and dissemination features is proposed, along with a four-level early warning mechanism. Finally, the effectiveness and practicality of the system are verified using real Weibo data. This system provides technical support for online public opinion governance.

Xia Jiayue, Du Shiping, Lang Yuxin, Ma Yuechen, Zhang Yu, Fan Yiqi. Design and Implementation of a Deep Learning-Based Intelligent Analysis and Early Warning System for Weibo Social Hot Topics. International Journal of Frontiers in Engineering Technology (2026), Vol. 8, Issue 2: 30-36. https://doi.org/10.25236/IJFET.2026.080205.

[1] CHEN T T, XU R F, HE Y L, et al. A sentiment analysis method for microblog based on BERT and BiLSTM-Attention[C]//Proceedings of 2020 IEEE International Conference on Progress in Informatics and Computing (PIC). Shanghai: IEEE, 2020: 50-55. 

[2] LI Y Z, XIE Y C, WANG X F, et al. Early warning of public opinion on social networks based on deep learning[J]. IEEE Access, 2020, 8: 2098-2009. 

[3] Zhang Shih-hao, Li Xin, Zhao Xin-yang, et al. Weibo public opinion heat analysis and prediction model integrating BERT and X-means algorithm [J]. Computer Applications, 2025, 45(10): 3138-3145.

[4] He Yuanzheng, You Huanxiang, Liu Pingyu, et al. Evolution mechanism of Weibo hot topics and public opinion guidance strategy under sudden public events - based on deep learning and sentiment analysis [R]. National Undergraduate Innovation Training Program, 2025.

[5] Lin Shaofu, Ren Dongliang. Sentiment early warning method based on public figure public opinion knowledge graph and XGBoost multi-feature fusion: CN112905800B [P]. 2025-01-14.

[6] Sarfraz A, Ahmad A, Zeshan F. Synews: a synergy-based rumor verification system[J]. Social Network Analysis and Mining, 2024, 14: 57.

[7] Cheng Zhecheng. Research and implementation of Weibo public opinion monitoring model based on deep learning [J]. Computer Era, 2023(11): 124-126,130.

[8] Zhu Y, Wang G, Huang X, et al. Enhancing real-time rumor detection on Weibo through user and content feature integration[J]. Computing and Informatics, 2025, 44(5): 1123-1140.

[9] Cheng Weidong. Early prediction and empirical study of network public opinion information heat based on deep learning [J]. Information Science, 2024, 42(2): 148-156.

[10] Li Haojie. Multimodal Brand Tracking and Event Detection on Social Media [R]. Dalian University of Technology, 2025.