GLAD
Related work
point-wise data: features of an individual
pair-wise: spcial ties
2 stages: group discovery and anomaly detection
static graph
MGM:
Latent Dirichlet Allocation(LDA): text processing
word–>topic–>document
group: mix of Gaussian distributed topics; exist the best mixture rate
discriminative approache:
kernel embedding of probabilistic distributions and generalizations OCSVM from point to group:
probablistic distribution —map(kernel)—>probability measure space
ATD: Anomalous Topic Discovery in High Dimensional Discrete Data
文本文档中的异常主题检测问题
非典型模式仅表现在超高维特征空间的一小(显著)子集的情况: 单独的AD技术和使用所有特征检测异常的技术通常不能检测到这样的异常
该方法可以集体检测这些实例,发现它们所表现出的共享异常模式,并识别显著特征的子集