Research On Data Analytics-oriented Exploratory Service Composition Recommendation Technology

The advent of the Internet of Things era has generated massive amounts of data,and data analysis of it using machine learning can extract relevant information from historical experiences to make them generate knowledge and use them to make new decisions.In the field of data analysis,it is a very complicated and time-consuming task for technicians to build suitable machine learning models based on specified data sets.In this process,it is necessary to select suitable algorithms at different stages and continuously adjust and optimize algorithm parameters,which often requires several explorations to get final results and requires high professional background for technicians.To address the problems of difficult algorithm selection and parameter optimization in the field of data analysis,this paper proposes an exploratory service recommendation method for data analysis,which recommends data analysis services for users based on data features and service association relationships,assists users in selecting better algorithms,and improves data analysis efficiency and machine learning model performance.The main work of this paper includes:Firstly,an exploratory service combination recommendation method for data analysis is proposed,which is divided into a user exploration part and a system support part.The user exploration part supports flexible service combination with end-user participation,which is used to cope with the problems of uncertain data analysis algorithm and uncertain logical relationship between services during the construction of exploratory data analysis process,and is responsible for providing support to the user exploratory construction of data analysis process,where the user selects data services,configures data analysis tasks,and gets a set of service recommendation list through service recommendation,and after multiple explorations,gets The system support part is responsible for providing service recommendations in the data analysis process and completing the execution tasks of the data analysis process.Secondly,a service recommendation algorithm combining data features and service associations is proposed,which is divided into service association mining algorithm,data analysis process generation algorithm and single-step service recommendation algorithm by steps.Among them,the service association mining algorithm can mine common service associations from historical data analysis processes;the data analysis process generation algorithm integrates data features and service associations to generate a set of data analysis processes;the single-step service recommendation algorithm extracts a set of data analysis services from data analysis processes based on performance and time-consuming factors as recommendation results.Thirdly,an exploratory service combination recommendation system for data analysis is designed and implemented.The system is mainly divided into five parts:service library,exploratory data analysis environment,service recommendation module,service association mining module and execution engine.The user selects data services from the service library,configures the data analysis tasks,constructs the data analysis process exploratively with the assistance of the service recommendation module,and finally gets the final results through the execution engine;the service association mining module is responsible for providing the recommendation basis for the service recommendation module.Experiment 1 compares the performance level and time consumption of data analysis processes generated by different data analysis process generation methods.The results show that compared with the mainstream Auto ML tools Autostacker,TPOT,and Alpha D3 M,the method used in this paper greatly reduces the data analysis process generation time in terms of time consumption,shortening the total time consumption to minutes.In terms of performance,it also achieves a better level of performance among the data analysis processes generated by current Auto ML tools.Experiment 2 compares the performance of the data analysis process obtained using the service recommendation method proposed in this paper with the data analysis process generated by other mainstream Auto ML tools,and then compares the service recommendation effect with TPOT,and the results show that,compared with the mainstream tools TPOT,Autostacker,and Alpha D3 M,the data analysis process constructed using the service recommendation method proposed in this paper The results show that,compared with TPOT,Autostacker,and Alpha D3 M,the data analysis process constructed by the service recommendation method proposed in this paper can achieve the medium or higher performance level of the mainstream Auto ML tools;in terms of the accuracy and recall of the service recommendation results,the performance of the service recommendation method proposed in this paper is also better than that of TPOT.