Research And Implementation Of Cloud Computing Based Anomaly Detection Techniques For Big Graphs

As a kind of common data structure,graph is increasingly being used for modeling scientific data.With the advent of the era of big data,the scale of of all kinds of graph data has become increasingly complex and huge.Graph data's powerful representation ability makes it can express a wealth of information on a variety of occasions.Detecting anomaly information in such huge graph data information has important practical significance,and has become a hot topic in today's academia and industry world.At present,the anomaly on the graph industry also does not have a unified complete definition,this thesis defines the anomaly on the graph as the its characteristics significantly different from most of the data which also represents unusual or interesting objects.Due to the complexity and diversity of the problems,the anomaly detection problem still does not have a complete set of theoretical solution.How to detect the anomaly according to different types of graphs of abnormal information,and how to solve the problem with the background of big data analysis of abnormal information in below will be the main problem in this thesis.With the further study of the characteristics of the network,people found that many of the actual networks all have a common nature,namely the community structure.That is to say,many real networks are composed of multiple community structure,community within the relatively close connection of the vertex,communication between the community and the community is relatively sparse.According to the characteristics of actual network,this thesis consider to use graph data of the correlation between vertices in anomaly detection,the vertex at the network which cross communities or outliers is likely to be abnormal vertices on the graph.In view of the above problem,this article in view of the different types of graph data,in make full use of the structure information of graph data and attribute information,put forward different anomaly detection strategy,main contributions are as follows:(1)For the plain graph,we put forward the main idea of detection is judging the similarity between the query vertex neighbors,then according to the similarity score between vertices for query vertex normal score.(2)For the attribute graph,this article adopts the way of finding community outliers to anomaly detection,the developed method in the detection of outliers often only focus on the structure of the graph data or attribute information,this article application based on overlapping community discovery algorithm to detect outliers,at the same time take into consideration of the vertex attribute information to achieve better results.This anomaly detection algorithm based on the model of the BSP.The experimental results show that in large-scale graph data,the algorithm can be effective for anomaly detection.