基于多模态特征的暗链标题检测方法

doi:10.3969/j.issn.1671-1122.2026.03.012

摘要/Abstract

摘要：

随着网页篡改植入暗链现象的愈演愈烈以及自动化检出方法的普及，暗链标题植入已成为危害网络安全的重要因素之一。当前，攻击者常采用形近字、干扰符号、表情文字等手段进行伪装，这对基于单模态自然语言处理的检测技术构成了挑战。针对这一问题，文章提出基于混合特征的多模态检测方法。该方法首先利用BERT与ResNet分别提取标题文本的语义特征与图像特征，随后通过门函数和多头注意力方法对特征进行深度融合，进而实现对暗链标题的分类。实验结果表明，在评测数据集上，所提方法的识别准确率达到0.966，较基准方法提升了约1个百分点，这表明图像特征可以有效弥补文本特征在应对标题伪装时的不足。

关键词: 暗链标题检测, BERT, ResNet, 多模态特征融合

Abstract:

As the growing phenomenon of web page tampering with implanted hidden links, and the popularity of automatic detection methods, hidden link headline implantation has become one of the important factors endangering network security. Currently, the detection rate of unimodal, natural language processing-based detection techniques gradually decreases as hidden link attackers adopt disguises such as morphological close characters, interference symbols, and emoticons. To address this problem, this paper proposed a multimodal detection method based on image features and text features. The proposed method first extracted the semantic features and image features of the headline text with BERT and ResNet respectively, and then based on the gate function and multi-headed attention methods, the features were deeply fused to achieve the classification of hidden link headlines. Experimental results on the evaluation dataset show that the recognition accuracy of the proposed method can reach 0.966, which is about 1 percentage points higher than that of the benchmark method. This indicates that the image features can effectively overcome the shortage that text features cannot cope with the problem of headline disguise.

Key words: hidden link headline detection, BERT, ResNet, multi-modal feature fusion

中图分类号:

TP309

印杰, 刘家银, 黄肖宇, 兰浩良, 谢文伟. 基于多模态特征的暗链标题检测方法[J]. 信息网络安全, 2026, 26(3): 462-470.

YIN Jie, LIU Jiayin, HUANG Xiaoyu, LAN Haoliang, XIE Wenwei. Hidden Link Headline Detection Method Based on Multi-Modal Features[J]. Netinfo Security, 2026, 26(3): 462-470.

图/表 10

表1

图1

图2

表2

图3

表3

表4

表5

图4

表6

参考文献 20

[1]	China Internet Network Information Center. The 49th China Statistical Report on Internet Development[EB/OL]. (2022-02-25)[2025-08-03]. https://www.cnnic.net.cn/n4/2022/0401/c88-1131.html.
	中国互联网络信息中心. 第49次中国互联网络发展状况统计报告[EB/OL]. (2022-02-25)[2025-08-03]. https://www.cnnic.net.cn/n4/2022/0401/c88-1131.html.
[2]	YIN Jie, JIANG Yuxiang, NIU Bowei, et al. Remote Detection of Web Page Tampering Based on Deep Learning[J]. Journal of Nanjing University of Science and Technology, 2020, 44(1): 49-54.
	印杰, 蒋宇翔, 牛博威, 等. 基于深度学习的网页篡改远程检测研究[J]. 南京理工大学学报, 2020, 44(1): 49-54.
[3]	BARTOLI A, DAVANZO G, MEDVET E. A Framework for Large-Scale Detection of Web Site Defacements[J]. ACM Transactions on Internet Technology, 2010, 10(3): 1-37.
[4]	ZHOU Wenyi, GU Xubo, SHI Yong, et al. Detection Method for Hidden Hyperlink Based on Machine Learning[J]. Computer Engineering, 2018, 44(10): 22-27. doi: 10.19678/j.issn.1000-3428. 0051189
	周文怡, 顾徐波, 施勇, 等. 基于机器学习的网页暗链检测方法[J]. 计算机工程, 2018, 44(10): 22-27.
[5]	DAVANZO G, MEDVET E, BARTOLI A. Anomaly Detection Techniques for a Web Defacement Monitoring Service[J]. Expert Systems with Applications, 2011, 38(10): 12521-12530. doi: 10.1016/j.eswa.2011.04.038 URL
[6]	MENG Chijie, WANG Wei, GENG Guanggang. Internet Hidden Hyperlinks Detection Based on Statistical Machine Learning[J]. Application Research of Computers, 2015, 32(9): 2779-2783.
	孟池洁, 王伟, 耿光刚. 基于统计机器学习的互联网暗链检测方法[J]. 计算机应用研究, 2015, 32(9): 2779-2783.
[7]	LIM M, ABDULLAH A, JHANJHI N, et al. Hidden Link Prediction in Criminal Networks Using the Deep Reinforcement Learning Technique[EB/OL]. (2019-01-11)[2025-08-03]. https://doi.org/10.3390/computers8010008.
[8]	SARAVANOU A, KATAKIS I, VALKANAS G, et al. Revealing the Hidden Links in Content Networks: An Application to Event Discovery[C]// ACM. The 2017 ACM on Conference on Information and Knowledge Management. New York: ACM, 2017: 2283-2286.
[9]	LIM M, ABDULLAH A, JHANJHI N Z. Performance Optimization of Criminal Network Hidden Link Prediction Model with Deep Reinforcement Learning[J]. Journal of King Saud University-Computer and Information Sciences, 2021, 33(10): 1202-1210. doi: 10.1016/j.jksuci.2019.07.010 URL
[10]	PANCHAL B Y, CHAUHAN G, PANCHAL S R, et al. An Investigation on Feature and Text Extraction from Images Using Image Recognition in Android[J]. Materials Today: Proceedings, 2022, 51: 798-802. doi: 10.1016/j.matpr.2021.06.237 URL
[11]	HE Xiaomin, WANG Jin, HE Yueying, et al. A Deep Learning Approach for Website Fingerprinting Attack[C]// IEEE. 2018 IEEE 4th International Conference on Computer and Communications (ICCC). New York: IEEE, 2019: 1419-1423.
[12]	BHAT S, LU D, KWON A, et al. Var-CNN: A Data-Efficient Website Fingerprinting Attack Based on Deep Learning[J]. Proceedings on Privacy Enhancing Technologies, 2019(4): 292-310.
[13]	DEVLIN J, CHANG Mingwei, LEE K, et al. BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding[C]//ACL. The 2019 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies. New York: ACL, 2019: 4171-4186.
[14]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep Residual Learning for Image Recognition[C]// IEEE. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2016: 770-778.
[15]	LEI Chengmin, MU Shaomin, SUN Wenjie, et al. Image Recognition of Peach Pests Based on Multi-Scale Attention Residual Network[J]. Journal of Shandong Agricultural University (Natural Science Edition), 2022, 53(2): 253-258.
	类成敏, 牟少敏, 孙文杰, 等. 基于多尺度注意力残差网络的桃树害虫图像识别[J]. 山东农业大学学报(自然科学版), 2022, 53(2): 253-258.
[16]	CAI Yuhan. Image Classification Algorithm Based on Convolutional Self-Coding and Residual Network[J]. Information Technology and Informatization, 2022(7): 73-76.
	蔡雨涵. 基于卷积自编码和残差网络的图像分类算法[J]. 信息技术与信息化, 2022(7): 73-76.
[17]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is All You Need[C]// ACM. The 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 6000-6010.
[18]	NATHANI D, CHAUHAN J, SHARMA C, et al. Learning Attention-Based Embeddings for Relation Prediction in Knowledge Graphs[C]// ACL. The 57th Annual Meeting of the Association for Computational Linguistics. New York: ACL, 2019: 4710-4723.
[19]	REDMON J, FARHADI A. YOLOv3:An Incremental Improvemen[EB/OL]. (2018-04-08)[2025-08-03]. https://doi.org/10.48550/arXiv.1804.02767.
[20]	KIM W, LEE J, PARK E, et al. Advanced Mechanism for Reducing False Alarm Rate in Web Page Defacement Detection[EB/OL]. (2006-01-01)[2025-08-03]. https://www.researchgate.net/publication/228953432_Advanced_Mechanism_for_Reducing_False_Alarm_Rate_in_Web_Page_Defacement_Detection.

类型编号	标题类型	样本数量
0	正常网页标题	11938
1	赌博类网站标题	2486
2	情色类网站标题	2023
3	其他异常标题	486

层名称	输出尺寸	f 0层
Conv1	112×112	7×7, 64, stride 2 3×3max pool, stride 2
Conv2_x	28×28	$\left[ \begin{matrix} 1\times 1 & 64 \\ 3\times 3 & 64 \\ 1\times 1 & 256 \\ \end{matrix} \right]\times 3$
Conv3_x	28×28	$\left[ \begin{matrix} 1\times 1 & 128 \\ 3\times 3 & 128 \\ 1\times 1 & 512 \\ \end{matrix} \right]\times 4$
Conv4_x	14×14	$\left[ \begin{matrix} 1\times 1 & 256 \\ 3\times 3 & 256 \\ 1\times 1 & 1024 \\ \end{matrix} \right]\times 6$
Conv5_x	7×7	$\left[ \begin{matrix} 1\times 1 & 512 \\ 3\times 3 & 512 \\ 1\times 1 & 2048 \\ \end{matrix} \right]\times 3$
Fc layer	1×1	Average_pool, 1000-d fc

参数名	参数值
文本维度	32
图像维度	224
BERT多头注意力头的个数/个	8
BERT隐藏层层数/层	4
[CLS]张量维度	256
多头注意力个数/个	4
优化算法	Adam
批大小	8
初始学习率	0.001
最终学习率	0.0001

方法	Accuracy	Precision	Recall	F1
n-gram（二元）	0.957	0.928	0.926	0.927
CART	0.918	0.844	0.886	0.864
GBDT	0.942	0.929	0.870	0.899
RF	0.952	0.918	0.920	0.919
单独使用BERT	0.956	0.916	0.936	0.926
单独使用ResNet	0.845	0.773	0.674	0.720
本文方法	0.966	0.919	0.970	0.944

算法	Accuracy	Precision	Recall	F1
+简单融合模型	0.954	0.919	0.926	0.922
+门函数	0.958	0.910	0.950	0.930
+注意力机制	0.956	0.913	0.942	0.927
本文方法	0.962	0.933	0.940	0.936