假新聞檢測已成為數位時代的一個重大挑戰，需要精細的方法來區分真實與誤導性資訊。傳統方法往往僅依賴文字內容分析，但這可能無法捕捉到準確檢測所需要的複雜語義關係與脈絡資訊。在本論文中，我們提出KAGNet (Knowledge-Augmented Graph Networks)，這是一種新穎的雙視圖圖神經網路架構，結合了局部與全域結構資訊，以及由大型語言模型(LLMs)生成的外部知識，以提升假新聞檢測的性能與穩定度。
我們的方法構建了一個異質圖，包含新聞節點、由LLM生成知識所衍生的含有外部知識的實體節點，以及主題節點。該模型採用雙視角架構，包含局部與全域視圖，利用不同層數的圖神經網路來捕捉不同尺度的資訊。一項關鍵創新是逐層注意力機制(layer-wise attention)，能夠動態地為不同網路層分配權重，使模型能夠自適應地為每個節點選擇最相關的資訊。此外，我們引入了相關與不相關知識的整合，讓模型能同時從新聞與實體之間的不同關係中學習。
為了確保有效學習，我們設計了門控多樣性正則化(gating diversity regularization)，鼓勵局部與全域視圖學習互補的表示，避免冗餘並增強模型的表達能力。訓練過程結合監督式學習與一致性正則化，以維持不同視角間的連貫性。
在四個基準資料集上的大量實驗顯示，KAGNet明顯優於現有的基準方法。我們的模型在GossipCop 上達到86.65%的準確率與76.21%的F1-score，在PolitiFact 上達到91.02% 的準確率與90.29%的F1-score。消融實驗證實了各模組的重要性，特別是LLM生成外部知識以及引入相關與不相關知識的整合。此外，穩健性分析顯示 KAGNet 在雜訊環境與訓練資料有限的情況下，仍能保持優異的效能。

Fake news detection has become a critical challenge in the digital age, requiring sophisticated approaches to distinguish between authentic and misleading information. Traditional methods often rely solely on textual content analysis, which may fail to capture the complex semantic relationships and contextual information necessary for accurate detection. In this thesis, we propose KAGNet (Knowledge-Augmented Graph Networks), a novel multi-view graph neural network architecture that leverages both local and global structural information along with external knowledge generated by Large Language Models (LLMs) to enhance fake news detection performance.
Our approach constructs a heterogeneous graph comprising news nodes, entity nodes derived from LLM-generated knowledge, and topic nodes. The model employs a dual-perspective architecture with local and global views, utilizing different numbers of Graph Neural Network layers to capture information at various scales. A key innovation is the layer-wise attention mechanism that dynamically assigns weights to different network layers, allowing the model to adaptively select the most relevant information for each node. Additionally, we introduce relevant and irrelevant knowledge integration, where the model learns from both matching and non-matching relationships between news and entities.
To ensure effective learning, we implement gating diversity regularization that encourages the local and global views to learn complementary representations, preventing redundancy while enhancing the model’s expressive power. The training process combines supervised learning with consistency regularization to maintain coherence between different perspectives.
Extensive experiments on four benchmark datasets demonstrate that KAGNet significantly outperforms existing baseline methods. Our model achieves 86.65% accuracy and 76.21% F1-score on GossipCop, and 91.02% accuracy and 90.29% F1-score on PolitiFact. Ablation studies confirm the importance of each component, particularly the integration of LLM-generated knowledge and the multi-view architecture. Furthermore, robustness analysis shows that KAGNet maintains superior performance under noisy conditions and limited training data scenarios.

[1] David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver,and Colin Raffel. Mixmatch: A holistic approach to semi-supervised learning, 2019.
[2] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. In Jill Burstein, Christy Doran, and Thamar Solorio, editors, Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, Min-neapolis, Minnesota, June 2019. Association for Computational Linguistics.
[3] Maarten Grootendorst. Bertopic: Neural topic modeling with a class-based tf-idf pro-cedure, 2022.
[4] Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. On calibration of modern neural networks, 2017.
[5] William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs, 2018.
[6] Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems, 43(2):1–55, January 2025.
[7] Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations, 2017.
[8] Alfirna Rizqi Lahitani, Adhistya Erna Permanasari, and Noor Akhmad Setiawan. Cosine similarity to determine similarity measure: Study case in online essay assessment. In 2016 4th International Conference on Cyber and IT Service Management, pages 1–6, 2016.
[9] David M. J. Lazer, Matthew A. Baum, Yochai Benkler, Adam J. Berinsky, Kelly M. Greenhill, Filippo Menczer, Miriam J. Metzger, Brendan Nyhan, Gordon Pennycook, David Rothschild, Michael Schudson, Steven A. Sloman, Cass R. Sunstein, Emily A. Thorson, Duncan J. Watts, and Jonathan L. Zittrain. The science of fake news. Science, 359(6380):1094–1096, 2018.
[10] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. Retrieval-augmented generation for knowledge-intensive nlp tasks, 2021.
[11] Xiaoxiao Ma, Yuchen Zhang, Kaize Ding, Jian Yang, Jia Wu, and Hao Fan. On fake news detection with LLM enhanced semantics mining. In Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen, editors, Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 508–521, Miami, Florida, USA, November 2024. Association for Computational Linguistics.
[12] Qiong Nan, Qiang Sheng, Juan Cao, Beizhe Hu, Danding Wang, and Jintao Li. Let silence speak: Enhancing fake news detection with generated comments from large language models. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM ＇24, page 1732–1742. ACM, October 2024.
[13] Fernando Perez-Cruz. Kullback-leibler divergence estimation of continuous distributions. In 2008 IEEE International Symposium on Information Theory, pages 16661670, 2008.
[14] Nils Reimers and Iryna Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks, 2019.
[15] Claude Sammut and Geoffrey I. Webb, editors. TF–IDF, pages 986–987. Springer US, Boston, MA, 2010.
[16] Michael Schlichtkrull, Thomas N. Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph convolutional networks, 2017.
[17] Kai Shu, Deepak Mahudeswaran, Suhang Wang, Dongwon Lee, and Huan Liu. Fakenewsnet: A data repository with news content, social context and spatialtemporal information for studying fake news on social media, 2019.
[18] subho. Fake news detection challenge kdd 2020. https://kaggle.com/competitions/fakenewskdd2020, 2020. Kaggle.
[19] T. Thanapalasingam, L. van Berkel, P. Bloem, and P. Groth. Relational graph convolutional networks: a closer look. PeerJ Computer Science, 8:e1073, 2022.
[20] Camille Thibault, Gabrielle Péloquin-Skulski, Jacob-Junqi Tian, Florence Laflamme, Yuxiang Guan, Reihaneh Rabbany, Jean-François Godbout, and Kellin Pelrine. A guide to misinformation detection datasets, 2024.
[21] Soroush Vosoughi, Deb Roy, and Sinan Aral. The spread of true and false news online. Science, 359(6380):1146–1151, 2018.
[22] William Yang Wang. “liar, liar pants on fire”: A new benchmark dataset for fake news detection. In Regina Barzilay and Min-Yen Kan, editors, Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 422–426, Vancouver, Canada, July 2017. Association for Computational Linguistics.
[23] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models, 2023.
[24] Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 32(1):4–24, January 2021.
[25] Bingbing Xie, Xiaoxiao Ma, Jia Wu, Jian Yang, and Hao Fan. Knowledge graph enhanced heterogeneous graph neural network for fake news detection. IEEE Trans. on Consum. Electron., 70(1):2826–2837, February 2024.
[26] Bingbing Xie, Xiaoxiao Ma, Jia Wu, Jian Yang, Shan Xue, and Hao Fan. Heterogeneous graph neural network via knowledge relations for fake news detection. In Proceedings of the 35th International Conference on Scientific and Statistical Database Management, SSDBM ’23, New York, NY, USA, 2023. Association for Computing Machinery.
[27] Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629, 2022.
[28] Yuchen Zhang, Xiaoxiao Ma, Jia Wu, Jian Yang, and Hao Fan. Heterogeneous subgraph transformer for fake news detection. In Proceedings of the ACM Web Conference 2024, WWW ’24, page 1272–1282, New York, NY, USA, 2024. Association for Computing Machinery.
[29] Xinyi Zhou and Reza Zafarani. A survey of fake news: Fundamental theories, detection methods, and opportunities. ACM Computing Surveys, 53(5):1–40, September 2020.