The Research On Learning Rules For Complex Event Processing

CEP(Complex Event Processing)is a newly born technology based on event stream.It encapsulates different kind of data from a system as different events,through analyzing the connection between different events,using filtering,associating,aggregating and finally getting some high-level events from the low-level ones.Since CEP widely used in the real life,especially in the financial filed,it attracts more and more attention from some relevant enterprise and lots of researchers.During the process of developing a CEP system,the definition of the CEP rules is an inevitable step.In general,the domain expert should define the needed events at first,analyze the relationship between different events,and then write the CEP rules using Event Pattern Language.While under some conditions high-accuracy value of event's attributes are required.What's more,the CEP rules don't stay static all the time,some tiny but critical changes would occur as the system running in some special systems.Under those conditions,the CEP rules are changeable and diverse.Obviously it is a really hard even impossible job for the domain experts to define all the possible rules.That's why the research of learning CEP rules come into being.There are two main ways to learn the CEP rules.One is through extracting the history record of complex events,the other is updating the rules parameter on line.This paper introduces two main contributions,as following:(1)Proposing an algorithm called Constrain Modularized CEP Rules Learning Algorithm Based on Sequence Tree.It utilizes the special characteristic that all the positive traces(those history record having composite event occurred)share almost the same primitive events and the sequence among them,and build a sequence tree using the sequential relationship between different primitive events for combining the same prefix among positive traces and avoiding unnecessary computing.In the end of the paper we design a set of experiments,and the experimental results validate the thought that using the sequence tree to combining the positive trace is feasibly and effectively.(2)Proposing an algorithm called CEP Rules Learning Based on Shapelets.It imitates the idea that mining the most discriminative subsequence of time series in the field of Early Classification of Time Series(ECTS)according to the very many similarities between CEP and ECTS.Respectively design the experiments for the univariate primitive event rules learning for CEP and the multivariate primitive event rules learning for CEP using the actual dataset.The experimental results indicate the algorithm is effective.