Supporting Cohort Query Processing Inside A Relational Database

Internet applications often produce massive data about users' behaviors.People expect to find the trend of users' behaviors from the generated big data,so as to make better business decisions.Social scientists have developed cohort analysis to analyze the impact of age on human behaviors in a changing social environment.With cohort analysis,behaviors' trend research is divided into three steps:1)to divide users into different cohorts based on the birth action:2)to calculate users' age in each record of users' behaviors;3)to calculate aggregate behavior indexes for each(cohort,age)set.However,when analyzing the users' behaviors data generated by Internet application,the classical cohort analysis technology has encountered two difficulties.First,social scientists usually analyze the whole small data set.Therefore.they do not provide the technology of cohort analysis for extracting the part.Second,they only use birth time to distinguish user cohorts.because it is a key attribute of social change.However,in Internet applications,other birth feature of the users is also an important attribute for determining the cohort.Classical cohort analysis technique lacks the general way of defining c.ohorts.Further,in the traditional database system,classical cohort analysis task is hard to use SQL to describe and executes with low efficiency.In this paper,a generalized cohort analysis technique is proposed and the corresponding cohort query processing technology is implemented on the relational database.Three new cohort query operators are proposed to solve above two problems.Based on the proposed new operators,data analysts can describe a complex cohort analysis task as a simple cohort query.Finally,a cohort query processing technology based on relational database is proposed,which converts the proposed cohort query to SQL to execute and optimizes it appropriately.The paper validates the usability of the proposed cohort query through the example comparison,and proves the performance of the proposed cohort query processing technology through the experiment.