融合BiLSTM与CNN的推特黑灰产分类模型

doi:10.3778/j.issn.1002-8331.2308-0220

摘要/Abstract

摘要： 当前推特等国外社交平台，已成为从事网络黑灰产犯罪不可或缺的工具，对推特上黑灰产账号进行发现、检测和分类对于打击网络犯罪、维护社会稳定具有重大意义。现有的推文分类模型双向长短时记忆网络（bi-directional long short-term memory，BiLSTM）可以学习推文的上下文信息，却无法学习局部关键信息，卷积神经网络（convolution neural network，CNN）模型可以学习推文的局部关键信息，却无法学习推文的上下文信息。结合BiLSTM与CNN两种模型的优势，提出了BiLSTM-CNN推文分类模型，该模型将推文进行向量化后，输入BiLSTM模型学习推文的上下文信息，再在BiLSTM模型后引入CNN层，进行局部特征的提取，最后使用全连接层将经过池化的特征连接在一起，并应用softmax函数进行四分类。模型在自主构建的中文推特黑灰产推文数据集上进行实验，并使用TextCNN、TextRNN、TextRCNN三种分类模型作为对比实验，实验结果显示，所提的BiLSTM-CNN推文分类模型在对四类推文进行分类的宏准确率为98.32%，明显高于TextCNN、TextRNN和TextRCNN三种模型的准确率。

关键词: 文本分类, 双向长短期记忆网络（BiLSTM）, 卷积神经网络（CNN）, 黑灰产, 推特

Abstract: Currently, foreign social platforms, such as Twitter, have become indispensable tools for engaging in cyber black and gray crime, and the discovery, detection and classification of black and gray accounts on Twitter are of great significance for combating cyber crime and maintaining social stability. The existing tweet classification model bidirectional long short-term memory (BiLSTM) can learn the contextual information of tweets but cannot learn the local key information, and the convolution neural network (CNN) model can learn the local key information of tweets but cannot learn the contextual information of tweets. This paper combines the advantages of BiLSTM and CNN models and proposes BiLSTM-CNN tweet classification model, which vectorizes the tweets, inputs them into BiLSTM model to learn the contextual information of the tweets, and then introduces a CNN layer after the BiLSTM model for the extraction of local features, and finally uses a fully connected layer to connect the pooled features together, and applies the softmax function for quadruple classification. The model is experimented on the independently constructed Chinese Twitter black and gray tweets dataset, and three classification models, TextCNN, TextRNN, and TextRCNN, are used as the comparison experiments, and the experimental results show that the proposed BiLSTM-CNN tweets classification model of this paper has a macro-accuracy of 98.32% in classifying the four types of tweets, which is significantly higher than that of TextCNN, TextRNN and TextRCNN three models’ accuracy.

Key words: text classification, bidirectional long short-term memory (BiLSTM), convolutional neural network (CNN), black market, Twitter

朱恩德, 王威, 高见. 融合BiLSTM与CNN的推特黑灰产分类模型[J]. 计算机工程与应用, 2025, 61(1): 186-195.

ZHU Ende, WANG Wei, GAO Jian. Twitter Black Market Accounts Classification Model Incorporating BiLSTM and CNN[J]. Computer Engineering and Applications, 2025, 61(1): 186-195.

参考文献

[1] 刘宪权. 网络黑灰产上游犯罪的刑法规制[J]. 国家检察官学院学报, 2021, 29(1): 3-17.
LIU X Q. Criminal regulation on the upstream crimes of the cyber underground industry chain[J]. Journal of National Prosecutors College, 2021, 29(1): 3-17.
[2] HRIDOY S D, VISHAL R D, ADITYA A, et al. Detecting fake retweeters using Hawkesprocess and topic modeling[J]. IEEE Transactions on Information Forensics and Security, 2020, 15: 2667-2678.
[3] HRIDOY S D, KARTIK A, TANMOY C. DECIFE: detecting collusive users involved in blackmarket following services on Twitter[C]//Proceedings of the 32nd ACM Conference on Hypertext and Social Media, 2021.
[4] 古倩. 基于特征向量构建的文本分类方法研究[D]. 西安: 西安理工大学, 2019.
GU Q. Research on text classification method based on feature vecgtor construction[D]. Xi’an: Xi’an University of Technology, 2019.
[5] 杨波. 新浪微博热点话题发现研究[D]. 乌鲁木齐: 新疆大学, 2019.
YANG B. Research on hot topic discovery of Sina microblog[D]. Urumqi: Xinjiang University, 2019.
[6] 易顺明, 易昊, 周国栋. 采用情感特征向量的Twitter情感分类方法研究[J]. 小型微型计算机系统, 2016, 37(11): 2454-2458.
YI S M, YI H, ZHOU G D. Twitter sentiment classification with sentimental feature vector[J]. Journal of Chinese Computer Systems, 2016, 37(11): 2454-2458.
[7] 胡馨月. Twitter事件检测中的语义和情感分析[D]. 成都: 电子科技大学, 2018.
HU X Y. Semantic and sentiment analysis in twitter event detection[D]. Chengdu: University of Electronic Science and Technology of China, 2018.
[8] 贺友程. 基于mBERT的东南亚小语种推特文本情感分析[J]. 电脑知识与技术, 2023, 19(1): 74-76.
HE Y C. Sentiment analysis of southeast Asian small language Twitter texts based on mBERT[J]. Computer Knowledge and Technology, 2023, 19(1): 74-76.
[9] 孙晓雨. 基于语义场景分析的文本表情分析方法研究[D]. 南京: 南京邮电大学, 2022.
SUN X Y. Research on text emoji analysis method based on semantic scene analysis[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2022.
[10] ARORA U, DUTTA H S, JOSHIET B, et al. Analyzing and detecting collusiveusers involved in blackmarket retweeting activities[J]. ACM Transactions on Intelligent Systems and Technology, 2020, 11(3): 1-24.
[11] MEHROTRA A, SARREDDY M, SINGH S. Detection of fake Twitter followers using graph centrality measures[C]//Proceedings of the 2016 2nd International Conference on Contemporary Computing and Informatics, 2016: 499-504.
[12] LEE K, MAHMUD J, CHEN J, et al. Who will retweet this? detecting strangers from twitter to retweet information[J]. ACM Transactions on Intelligent Systems and Technology, 2015, 6(3): 1-25.
[13] SHAH N, BEUTEL A, GALLAGHER B, et al. Spotting suspicious link behavior with fbox: an adversarial perspective[C]//Proceedings of the 2014 IEEE International Conference on Data Mining, 2014: 959-964.
[14] CRESCI S, PIETRO R D, PETROCCHI M, et al. A fake follower story: improving fake accounts detection on Twitter[R]. 2014.
[15] JIANG M, CUI P, BEUTEL A, et al. CatchSync: catching synchronized behavior in large directed graphs[C]//Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2014: 941-950.
[16] KWAK H, CHUN H, MOON S. Fragile online relation-ship: a first look at unfollow dynamics in twitter[C]//Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 2011: 1091-1100.
[17] ARORA U, PAKA W S, CHAKRABORTY T. Multitask learning for blackmarket tweet detection[C]//Proceedings of the 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, 2019: 127-130.
[18] DUTTA H S, AGGARWAL K, CHAKRABORTY T. DECIFE: detecting collusive users involved in blackmarket following services on Twitter[C]//Proceedings of the 32nd ACM Conference on Hypertext and Social Media, 2021.
[19] KIM Y. Convolutional neural networks for sentence classification[C]//Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, Doha, 2014: 1746-1751.
[20] 刘浩. 侵犯公民个人信息罪的法益构造及其规范解释[J]. 环球法律评论, 2023, 45(3): 163-179.
LIU H. The legal interest structure of the crime of infringing on citizens’ personal information and its normative interpretation[J]. Global Law Review, 2023, 45(3): 163-179.
[21] 张巍. 涉网络犯罪相关行为刑法规制研究[D]. 上海: 华东政法大学, 2015.
ZHANG W. Research into criminal regulation of network relevant crimes[D]. Shanghai: East China University of Political Science and Law, 2015.
[22] 班艺源. 少捕慎诉慎押下“两卡”类帮信罪的司法治理[J]. 政法学刊, 2022, 39(6): 14-22.
BAN Y Y. On the judicial administration of “two cards” crime of assisting cybercrime under the background of less arrest and prudent prosecution and custody[J]. Journal of Political Science and Law, 2022, 39(6): 14-22.
[23] LIU P, QIU X, HUANG X. Recurrent neural network for text classification with multi-task learning[C]//Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence, New York, 2016: 2873-2879.
[24] LAI S, XU L, LIU K, et al. Recurrent convolutional neural networks for text classification[C]//Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence, 2015: 2267-2273.