Small knowledge, big challenge! This article is participating in the creation activity of “Essential Tips for Programmers”
This article also participated in the “Digitalstar Project” to win a creative gift package and creative incentive money
By summarizing several articles on anomaly detection in attribute graphs in 19 years, the research target is unsupervised static attribute graphs. No other such articles have been found at the summit in recent years.
The first paper
Zong B, Song Q, Min M R, et al. Deep autoencoding gaussian mixture model for unsupervised anomaly detection[J]. 2018.
The previous work is to map the high dimensional vector of topology structure to the low dimensional vector, and then do density estimation. This is the first time that AE and GMM are optimized together.
Article 2:
Ding K, Li J, Bhanushali R, et al. Deep anomaly detection on attributed networks[C]//Proceedings of the 2019 SIAM International Conference on Data Mining. Society for Industrial and Applied Mathematics, 2019: 594-602.
After using GCN, attribute loss and structure loss are constructed respectively. Fractions were introduced.
The third article
Li Y, Huang X, Li J, et al. SpecAE: Spectral AutoEncoder for Anomaly Detection in Attributed Networks[C]//Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 2019: 2233-2236.
Background: Current methods are based on the assumption of isomorphism: connected nodes have similar node attributes. Therefore, topological relations can be introduced by minimizing the attribute distance of connected nodes. The existing GCN approach lets each node interact with its neighbors, but this approach smooths the attributes of the node and ignores global exceptions. The problems solved in this paper are 1. How to define exceptions and 2. How to solve the GCN over-smoth problem.
Abnormal definition
Global exception: The attributes of a node are different from those of the whole.
Community exception: The neighbors of node I and node I have different attributes.
AE is used for global exceptions and AE of GCN is used for community exceptions. When the hidden vector is put into the Gaussian mixture model, the one on the second side of the gaussian mixture model can be regarded as the outlier.
Summarize the current unsupervised anomaly detection methods: 1. Reconstruction loss based on AE 2. Based on density estimation 3. The combination of AE+ density estimation is applicable. Such as specAE. For now, these papers focus on how to define exceptions! How to better represent node characteristics.
Focus on machine learning and risk control algorithms, share data mining contests, machine learning, anti-cheating, recommendation systems and other technical articles. Come and study with me!