Aiming at the problems that the existing anomaly detection methods for directed event based on hypergraph representation learning cannot fully capture the anomaly patterns and are limited in use, an anomaly detection method for directed event using multi-scale contrastive learning (ADDE-MCL) is proposed. Firstly, events are modeled as directed hypergraphs. Secondly, a multi-scale comparative learning algorithm is designed to capture the anomaly patterns of directed events in terms of entity pair, entity set, head and tail set and event direction. Finally, the anomaly event scoring function is designed to complete the abnormal event detection. Experiments on 3 real datasets in different fields show that the proposed method achieves a better precision and area under the curve (AUC) compared with the baseline method, with an average improvement of 11.84% in precision and 15.22% in AUC.
鉴于超图中的边能够表达任意个节点的交互关系,近年来越来越多的研究者开始用超图来建模事件,利用超图表示学习,将图结构映射为低维稠密向量,用于异常检测。随着深度学习技术在超图表示学习中表现良好,基于超图的异常事件检测方法成为了当前的主流。异构超图变分自编码器(Heterogeneous Hypergraph Variational Autoencoder, HeteHG-VAE)[13]将异质信息网络映射为异构超图,并设计随机变分编码器来学习节点和超边的嵌入表示。Event2vec[14]将事件建模为异质超图并提出一种新的嵌入框架,通过保留事件内和事件间相似性来学习实体的嵌入表示。基于超图对比学习的异常事件检测(Abnormal Event Detection via Hypergraph Contrastive Learning, AEHCL)[15]使用超边对属性异质信息网络中包含的事件进行建模,并提出一种无监督的超图对比学习方法来捕获事件异常模式。双曲异构信息网络嵌入模型(Hyperbolic Heterogeneous Information Network Embedding Model, HHNE)[16]提出一个新的异构信息网络(Heterogeneous Information Network, HIN)嵌入方法,使用双曲空间的距离作为相似性度量。
上述方法关注的是无向事件的异常检测,针对有向事件的异常检测研究较少,有些有向超边的研究可用于异常检测。神经超链接预测器-D(Neural Hyperlink Predictor-D, NHP-D)[17]最早开展有向超图中的链路预测问题,将有向超边划分为头尾两个子超边,然后利用图卷积网络(Graph Convolutional Network, GCN)来学习头尾超边嵌入,最后用头尾超边嵌入向量的兼容性来计算方向评分,可用于异常事件的检测。有向超图链路预测的两阶段框架(Two-Stage Framework for Directed Hypergraph Link Prediction, TF-DHP)[18]设计一个基于BiLSTM的方向推理模块,分别从头尾两个方向对超边进行表示学习,最后通过softmax获得方向得分用于异常检测。这两种有向超图的研究借助超图表示学习来获得超图结构的潜在低维表示,不是直接针对异常检测本身来设计的,直接用于异常检测时性能有限。同时,这些方法使用了图神经网络的有监督学习,训练时需要数据标签,在难以获得数据标签的情况下难以应用。
针对上述问题,提出利用多尺度对比学习的有向事件异常检测方法(Anomaly Directed Event Detection Method Using Multi-scale Contrastive Learning, ADDE-MCL)。首先,将事件建模为有向超图;其次,针对有向事件中的实体对、实体集、头尾集和事件方向等方面的异常,设计多尺度对比学习算法捕获有向事件异常模式;最后,融合多尺度对比学习结果,设计异常事件评分函数完成有向事件异常检测。实验结果表明,相比基线方法,ADDE-MCL的异常检测性能在精确率和曲线下面积(Aarea Under the Curve, AUC)两个指标上均取得最优,召回率除一个数据集取得次优外,均最优。
WANGH B, ZHOUC, WUJ, et al. Deep structure learning for fraud detection[C]∥Proceedings of the 2018 IEEE International Conference on Data Mining. Piscataway, USA: IEEE, 2018:567-576.
[4]
ZHENGM Y, ZHOUC, WUJ, et al. FraudNE: a joint embedding approach for fraud detection[C]∥Proceedings of the 2018 International Joint Conference on Neural Networks. Piscataway, USA: IEEE, 2018. DOI: 10.1109/IJCNN.2018.8489585 .
[5]
DUAND S, TONGL L, LIY X, et al. AANE: anomaly aware network embedding for anomalous link detection[C]∥Proceedings of the 2020 IEEE International Conference on Data Mining. Piscataway, USA: IEEE, 2020:1002-1007.
[6]
GUIH, LIUJ L, TAOF B, et al. Embedding learning with events in heterogeneous information networks[J]. IEEE Transactions on Knowledge and Data Engineering, 2017,29(11):2428-2441.
[7]
CHENT, TANGL A, SUNY Z, et al. Entity embedding-based anomaly detection for heterogeneous categorical events[C]∥Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence. Washington, USA: AAAI Press, 2016:1396-1403.
[8]
HUB B, ZHANGZ Q, SHIC, et al. Cash-out user detection based on attributed heterogeneous information network with a hierarchical attention mechanism[J]. Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence. Palo Alto, USA: AAAI Press, 2019:946-953.
[9]
FANS H, SHIC, WANGX. Abnormal event detection via heterogeneous information network embedding[C]∥Proceedings of the 27th ACM International Conference on Information and Knowledge Management. New York, USA: ACM, 2018:1483-1486.
[10]
DONGY X, CHAWLAN V, SWAMIA. metapath2vec: scalable representation learning for heterogeneous networks[C]∥Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York, USA: ACM, 2017:135-144.
[11]
WANGX, JIH Y, SHIC, et al. Heterogeneous graph attention network[C]∥Proceedings of the World Wide Web Conference. New York, USA: ACM, 2019:2022-2032.
[12]
WANGX, LIUN, HANH, et al. Self-supervised heterogeneous graph neural network with co-contrastive learning[C]∥Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. New York, USA: ACM, 2021:1726-1736.
[13]
FANH Y, ZHANGF B, WEIY X, et al. Heterogeneous hypergraph variational autoencoder for link prediction[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022,44(8):4125-4138.
[14]
CHUY F, FENGC Y, GUOC L, et al. Event2vec: heterogeneous hypergraph embedding for event data[C]∥Proceedings of the 2018 IEEE International Conference on Data Mining Workshops. Piscataway, USA: IEEE, 2018:1022-1029.
[15]
YANB, YANGC, SHIC, et al. Abnormal event detection via hypergraph contrastive learning[C]∥Proceedings of the 2023 SIAM International Conference on Data Mining. Philadelphia, USA: Society for Industrial and Applied Mathematics, 2023:712-720.
[16]
WANGX, ZHANGY D, SHIC. Hyperbolic heterogeneous information network embedding[C]∥Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence. Washington, USA: AAAI Press, 2019:5337-5344.
[17]
YADATIN, NITINV, NIMISHAKAVIM, et al. NHP: neural hypergraph link prediction[C]∥Proceedings of the 29th ACM International Conference on Information & Knowledge Management. New York, USA: ACM, 2020:1705-1714.
[18]
XIAOG C, LIAOJ Z, TANZ, et al. A two-stage framework for directed hypergraph link prediction[J]. Mathematics, 2022,10(14):No.2372.
[19]
FENGY F, YOUH X, ZHANGZ Z, et al. Hypergraph neural networks[C]∥Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence. Palo Alto, USA: AAAI Press, 2019:3558-3565.
[20]
YADATIN, NIMISHAKAVIM, YADAVP, et al. HyperGCN: a new method of training graph convolutional networks on hypergraphs[C]∥Proceedings of the 33rd Conference on Neural Information Processing Systems. Piscataway, USA: IEEE, 2019:No.32.