An SBERT ‐ LDA Framework With FADC ‐Based Semantic Aggregation for Short‐Text Topic Modeling of Online Public Opinion on Energy Enterprises
Concurrency and Computation: Practice and Experience · 2026-08 · 38 (15)
DOI: 10.1002/cpe.70874 ↗Kurz gefasst
Fachartikel in der Zeitschrift Concurrency and Computation: Practice and Experience, online erschienen am 22.07.2026. Verfasst von Liu Chun und 4 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben.
Abstract (Originalsprache)
With the rapid growth of digital social media, identifying hot topics in online public opinion has become important for public opinion monitoring and risk management in the energy sector. Latent Dirichlet allocation (LDA) has been widely used for topic discovery on social media platforms such as Weibo. However, traditional LDA‐based methods often face challenges when applied to short texts because individual posts contain limited word co‐occurrence information. In addition, the bag‐of‐words representation used by LDA limits its ability to capture contextual semantic relationships. To address these problems, this study proposes an SBERT‐LDA framework for short‐text topic modeling of online public opinion related to energy enterprises. The framework uses sentence‐BERT (SBERT) to capture sentence‐level semantic similarity and introduces a fast adaptive density clustering (FADC) algorithm to construct semantically coherent and size‐controllable pseudo‐documents. These pseudo‐documents enrich the document‐level word co‐occurrence context available to LDA, thereby alleviating short‐text sparsity while indirectly incorporating semantic information into the LDA‐based topic modeling process. Dimensionality reduction is further applied to improve the efficiency of clustering high‐dimensional embeddings. Extensive experiments were conducted on a large‐scale Weibo dataset related to Chinese energy enterprises, with additional validation on the public TNEWS benchmark. The results show that the proposed framework generates coherent and interpretable topics while maintaining competitive topic diversity. Ablation studies, FADC analysis, and human‐in‐the‐loop evaluation further support the effectiveness, efficiency, and interpretability of the framework. Applied to energy‐enterprise‐related Weibo posts, the SBERT‐LDA framework identified five distinct hot topics, ranging from corporate finance and stock market trends to new energy technologies, oil and gas markets, and energy infrastructure, providing useful insights for strategic public opinion analysis.
Zitieren
APA
Chun, L., Wang, J., Zhao, Y., Wang, C., & Shangguan, J. (2026). An SBERT ‐ LDA Framework With FADC ‐Based Semantic Aggregation for Short‐Text Topic Modeling of Online Public Opinion on Energy Enterprises. Concurrency and Computation: Practice and Experience, 38(15). https://doi.org/10.1002/cpe.70874
Vancouver
Chun L, Wang J, Zhao Y, Wang C, Shangguan J. An SBERT ‐ LDA Framework With FADC ‐Based Semantic Aggregation for Short‐Text Topic Modeling of Online Public Opinion on Energy Enterprises. Concurrency and Computation: Practice and Experience. 2026;38(15). doi:10.1002/cpe.70874
BibTeX
@article{Chun2026,
author = {Chun, Liu and Wang, Jiakun and Zhao, Yibo and Wang, Chengwei and Shangguan, Jiaen},
title = {An SBERT ‐ LDA Framework With FADC ‐Based Semantic Aggregation for Short‐Text Topic Modeling of Online Public Opinion on Energy Enterprises},
journal = {Concurrency and Computation: Practice and Experience},
year = {2026},
volume = {38},
number = {15},
doi = {10.1002/cpe.70874}
}
EUVIMED