Event Modeling And Analysis Based On Bipartite Graph

In the computer field, event is the object of the system activity to record. Event reflects state changes dynamically, there are relationships between events, between event sources and between event and event source, relationships of events contain important information which can provide suggestions for other services. Current event analysis method contains log analysis and complex event processing. Log analysis tools analyze the system logs, user access logs, web logs, or others. Complex event processing extracts meaningful events in multiple event streams.However, neither logs nor the real-time streams can reflect the relations between events, or between event sources, or between event and event source directly. Moreover, with the increase of event number and producing speed, log analysis and complex event processing are not good at depth mining of event relationships in large-scale events.Graph is a common data structure, which can express the complex relationship between vertices naturally and intuitively. It has been widely used in social network, biological information, route planning and so on.Bipartite graph is a special graph structure, which contains two kinds of vertexes. Bipartite graph can be used in recommendation systems. Log analysis and complex event processing are not good at expressing and analyzing event relation but graph structure has the advantages. To reflect and analysis event relations intuitively and deeply, this paper proposed an event modeling and analysis method based on bipartite graph and implemented an event storage and analysis system.The research work of this paper is divided into the following three aspects:Firstly, an event modeling method based on bipartite graph is proposed. The method extended bipartite graph to a directed property graph. One event is divided into two parts: the event source and the event action which are modeled as the vertexes of the graph. The relationship between event source and event action is modeled as the edge.Secondly, an event analysis method based on SimRank is proposed.The method combines edge weights and vertex-centric process model to calculate the relationship between vertexes.Thirdly, an event storage and analysis system is designed and implemented. The system contains complete processes which are event modeling, event storage, event querying and event analysis. In this system, events are stored as eventsource-eventaction bipartite graph. And the event analysis method is used to analyze the correlation between the two event sources.At last, the event modeling and analysis method and the system are tested based on experiments. The experimental results show that the proposed modeling and analysis method based on bipartite graph has the effectiveness, and the event storage and analysis system can meet the expected demand in function and performance.