基于大模型数据增强的多粒度特征融合谣言检测

doi:10.19678/j.issn.1000-3428.0260022

摘要/Abstract

摘要： 随着网络和社交媒体的快速发展,信息的生成和传播速度达到了前所未有的水平,虚假信息、谣言及其他误导性内容充斥的现象愈加突出,这类问题已对社会治理秩序、和谐稳定构成重大威胁。在谣言检测中,谣言样本占比低导致数据不平衡,现有文本增强技术因缺乏谣言风格针对性、生成质量低,难以提升检测效果。同时,预训练语言模型虽然擅长捕捉文本全局依赖关系,却难以聚焦谣言关键局部特征。为此,提出一种基于大模型数据增强的多粒度特征融合的谣言检测框架。首先,提出融合谣言风格词典与大语言模型(LLM)的谣言生成方法,基于公开谣言数据集构建风格词典,以词典为约束指导LLM生成语义连贯且符合谣言风格的少数类样本,在缓解数据不平衡问题的同时保障增强样本质量;其次,所提多粒度上下文特征提取器(MGCFE),融合基于解耦注意力机制的预训练语言模型在全局依赖捕捉上的优势,与卷积子层对局部特征的聚焦能力,实现对谣言语义长距离逻辑关联与细粒度语言线索的同步捕捉,有效弥补此类预训练模型在局部关键特征捕捉上的固有局限。实验结果表明,该检测方法在BuzzFeed和PolitiFact数据集准确率分别达到82.24%、93.91%。

关键词: 深度学习, 大语言模型, 数据增强, 谣言检测, 多粒度特征融合, 社会治理

Abstract: With the rapid development of the internet and social media, the speed of information generation and dissemination has reached an unprecedented level. The proliferation of misinformation, rumors, and other misleading content has become increasingly prominent, posing significant threats to the order, harmony, and stability of social governance. In rumor detection, a low proportion of rumor samples leads to data imbalance, and existing text augmentation techniques struggle to enhance detection performance owing to their lack of specificity to rumor styles and low generation quality. Additionally, although pretrained language models capture global dependencies in text, they often fail to focus on the key local features of rumors. To address these challenges, this study proposes a rumor detection framework based on large-model data augmentation and multi-granularity feature fusion. First, a rumor generation method integrating a rumor-style lexicon and Large Language Model (LLM) is proposed. Based on publicly available rumor datasets, a style lexicon is constructed to guide LLM in generating semantically coherent and rumor-style consistent minority class samples. This approach alleviates data imbalance while ensuring the quality of the augmented samples. Second, this study introduces a Multi-Granularity Contextual Feature Extractor (MGCFE), which combines the strengths of pretrained language models with disentangled attention mechanisms to capture global dependencies and focus convolutional sublayers on local features. This enables the simultaneous capture of long-distance logical associations and fine-grained linguistic clues in rumor semantics, effectively mitigating the inherent limitations of pretrained models in capturing key local features. The experimental results demonstrate that the proposed detection method achieves accuracy rates of 82.24% and 93.91% for BuzzFeed and PolitiFact datasets, respectively.

Key words: deep learning, Large Language Model (LLM), data augmentation, rumor detection, multi-granularity feature fusion, social governance

中图分类号:

TP181

梁堉, 马佳妍, 胡晰远, 王子恒, 刘文, 彭天豪, 李莹. 基于大模型数据增强的多粒度特征融合谣言检测[J]. 计算机工程, 2026, 52(10): 390-406.

LIANG Yu, MA Jiayan, HU Xiyuan, WANG Ziheng, LIU Wen, PENG Tianhao, LI Ying. Multi-Granularity Feature Fusion for Rumor Detection Based on Large Model Data Augmentation[J]. Computer Engineering, 2026, 52(10): 390-406.

https://www.ecice06.com/CN/Y2026/V52/I10/390

参考文献

[1] ZHANG K W, YU J C, SHI H C, et al. Rumor detection with diverse counterfactual evidence[EB/OL].[2025-12-02]. https://arxiv.org/pdf/2307.09296.
[2] LIN H Z, MA J, CHEN L L, et al. Detect rumors in microblog posts for low-resource domains via adversarial contrastive learning[EB/OL].[2025-12-02]. https://arxiv.org/pdf/2204.08143.
[3] MA X X, ZHANG Y C, DING K Z, et al. On fake news detection with LLM enhanced semantics mining [C]//Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Florida, USA: Association for Computational Linguistics, 2024: 508-521.
[4] 马满福, 陈嘉豪, 李勇, 等. 基于改进GAT的多特征融合谣言检测模型MFLAN[J].计算机工程, 2025, 51(8):181-189. MA M F, CHEN J H, LI Y, et al. Multi-feature fusion rumor detection model MFLAN based on improved graph attention network[J]. Computer Engineering, 2025, 51(8): 181-189. (in Chinese)
[5] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[EB/OL].[2025-12-02]. https://arxiv.org/abs/1706.03762?ref=andrealarosa.org.
[6] YANG B S, TU Z P, WONG D K, et al. Modeling localness for self-attention networks[C]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels, USA: Association for Computational Linguistics, 2018: 4449-4458.
[7] KIM Y. Convolutional neural networks for sentence classification[C]//Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2014: 1746-1751.
[8] 吴菲, 陈嘉诚,杨俊,等. 基于自适应双域注意力网络的遥感图像重建[J]. 光电工程,2025,52(4): 240297. WU F, CHEN J C, YANG J, et al. Remote-sensing images reconstruction based on adaptive dual-domain attention network[J]. Opto-Electronic Engineering, 2025, 52(4): 240297. (in Chinese)
[9] KWON S, CHA M, JUNG K, et al. Prominent features of rumor propagation in online social media[C]//Proceedings of the 13th International Conference on Data Mining. Washington D.C., USA: IEEE Press, 2013:1103-1108.
[10] SHRESTHA A, SPEZZANO F. Textual characteristics of news title and body to detect fake news: a reproducibility study[C]//Proceedings of the Advances in Information Retrieval: the 43rd European Conference on IR Research. Berlin, Germany: Springer, 2021:120-133.
[11] SHRESTHA A, SPEZZANO F, GURUNATHAN I. Multi-modal analysis of misleading political news[C]//Proceedings of the Disinformation in Open Online Media: Second Multidisciplinary International Symposium. Berlin, Germany: Springer, 2020:261-276.
[12] MA J, GAO W, MITRA P, et al. Detecting rumors from microblogs with recurrent neural networks[C]// Proceedings of the 25th International Joint Conference on Artificial Intelligence. Palo Alto, USA: AAAI Press, 2016:3818-3824.
[13] HOJJAT S, SHARAN S, JOSEPH B, et al. Recent advances in recurrent neural networks[EB/OL]. [2025-12-02]. https://arxiv.org/pdf/1801.01078.
[14] 许旻辰, 屈丹, 司念文, 等. 社交媒体虚假信息检测技术研究综述[J].计算机工程, 2026,52(5):60-80. XU M C, QU D, SI N W, et al. A survey of the technologies for detecting disinformation in social media[J]. Computer Engineering, 2026,52(5):60-80. (in Chinese).
[15] SINGH J P, KUMAR A, RANA N P, et al. Attention-based LSTM network for rumor veracity estimation of Tweets[J]. Information Systems Frontiers, 2022, 24(2): 459-474.
[16] WANG Y Q, MA F J, JIN Z W, et al. EANN: event adversarial neural networks for multi-modal fake news detection[C]//Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York, USA: ACM Press, 2018: 849-857.
[17] SINGHAL S, SHAH R R, CHAKRABORTY T, et al. SpotFake: a multi-modal framework for fake news detection[C]//Proceedings of the 15th International Conference on Multimedia Big Data. Washington D.C., USA: IEEE Press, 2019:39-47.
[18] DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Minneapolis, Minnesota: Association for Computational Linguistics, 2019: 4171-4186.
[19] LIU Y H, OTT M, GOYAL N, et al. RoBERTa: a robustly optimized BERT pretraining approach[EB/OL].[2025-12-02]. https://arxiv.org/pdf/1907.11692.
[20] WEI J, ZOU K. EDA: easy data augmentation techniques for boosting performance on text classification tasks[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Hong Kong, China: Association for Computational Linguistics, 2019: 6382-6388.
[21] WANG Y, MA F Y, YANG Z Q, et al. Rumor detection with adaptive data augmentation and adversarial training[J]. Journal of Artificial Intelligence Research, 2025, 82: 1175-1204.
[22] RICO S, BARRY H, ALEXANDRA B. Improving neural machine translation models with monolingual data[C]//Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. Berlin, Germany: Association for Computational Linguistics, 2016: 86-96.
[23] HUA J H, CUI X D, LI X H, et al. Multimodal fake news detection through data augmentation-based contrastive learning[J]. Applied Soft Computing, 2023, 136(1): 110125-110133.
[24] AMJAD M, SIDOROV G, ZHILA A. Data augmentation using machine translation for fake news detection in the Urdu language[C]//Proceedings of the 12th Language Resources and Evaluation Conference. Marseille, France: European Language Resources Association, 2020: 2537-2542.
[25] SANH V, DEBUT L, CHAUMOND J, et al. DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter[EB/OL].[2025-12-02]. https://arxiv.org/pdf/1910.01108.
[26] SUYANTO S G. Synonyms-based augmentation to improve fake news detection using bidirectional LSTM[C]//Proceedings of the 8th International Conference on Information and Communication Technology. Washington D.C., USA: IEEE Press, 2020:1-5.
[27] BRANCO P, TORGO L, RIBEIRO R P. A survey of predictive modeling on imbalanced domains[J]. ACM Computing Surveys, 2016, 49(2): 31.
[28] HE H, BAI Y, GARCIA E A, et al. ADASYN: adaptive synthetic sampling approach for imbalanced learning [C]//Proceedings of the 2008 IEEE International Joint Conference on Neural Networks. Hong Kong, China: IEEE Press, 2008: 1322-1328.
[29] CATALTAS M, CICEKLI I, BAYKAN N A. Data augmentation for text classification using autoencoders[J]. IEEE Access, 2025, 13:161594-161604.
[30] NESHAEI S P, DAVIS R L, MEJIA-DOMENZAIN P, et al. Bridging the data gap: using LLMs to augment datasets for text classification[C]//Proceedings of the 18th International Conference on Educational Data Mining. Palermo, Italy: International Educational Data Mining Society, 2025: 119-132.
[31] 石宇, 于宁, 孙亚伟, 等. 基于元多任务提示学习的零样本谣言检测方法[J]. 北京邮电大学学报, 2024, 47(4): 77-82. SHI Y, YU N, SUN Y W, et al. A zero-shot rumor detection method based on meta multi-task prompt learning[J]. Journal of Beijing University of Posts and Telecommunications, 2024, 47(4): 77-82. (in Chinese)
[32] LI Y C, DING K Z, WANG J L, et al. Empowering large language models for textual data augmentation[EB/OL].[2025-12-02]. https://arxiv.org/pdf/2404.17642.
[33] 盛坤, 王中卿. 基于大语言模型和数据增强的通感隐喻分析[J]. 计算机应用, 2025, 45(3): 794-800. SHENG K, WANG Z Q. Synaesthetic metaphor analysis based on large language models and data augmentation[J]. Journal of Computer Applications, 2025, 45(3): 794-800. (in Chinese)
[34] SU J Y, CARDIE C, NAKOV P, et al. Adapting fake news detection to the era of large language models[EB/OL].[2025-12-02]. https://arxiv.org/pdf/2311.04917.
[35] HE P C, LIU X D, GAO J F, et al. DeBERTa: decoding-enhanced BERT with disentangled attention[EB/OL].[2025-12-02]. https://arxiv.org/pdf/2006.03654.
[36] POTTHAST M, KIESEL J, REINARTZ K, et al. A stylometric inquiry into hyperpartisan and fake news[EB/OL].[2025-12-02]. https://arxiv.org/pdf/1702.05638.
[37] LAI J Q, YANG X R, LUO W Y, et al. RumorLLM: a rumor large language model-based fake-news-detection data-augmentation approach[J]. Applied Sciences, 2024, 14(8): 3532.
[38] TORSHIZI A S, GHAZIKHANI A. Automatic Twitter rumor detection based on LSTM classifier[C]//Proceedings of High-Performance Computing and Big Data Analysis. Berlin, Germany: Springer, 2019: 291-300.
[39] ZENG Y R, DING X, CAI B B, et al. Exploring large language models for effective rumor detection on social media[C]//Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies. Albuquerque, New Mexico: Association for Computational Linguistics, 2025: 2537-2552.
[40] TIAN Z L, HUANG J Y, HE Z J, et al. LLM-based rumor detection via influence guided sample selection and game-based perspective analysis[C]//Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. Vienna, Austria: Association for Computational Linguistics, 2025: 28402-28414.
[41] HOLTZMAN A, JAN B, LI D, et al. The curious case of neural text degeneration[C]//Proceedings of the 8th International Conference on Learning Representations. New York, USA: [s.n], 2020: 95-99.
[42] BENGIO Y, DUCHARME R, VINCENT P, et al. A neural probabilistic language model[J]. The Journal of Machine Learning Research, 2003, 3: 1137-1155.
[43] HUTTO C J, GILBERT E. VADER: a parsimonious rule-based model for sentiment analysis of social media text[C]//Proceedings of the International AAAI Conference on Web and Social Media. Palo Alto, USA: AAAI Press, 2014: 216-225.

选择文件类型/文献管理软件名称

选择包含的内容