Index Selection Method Research Based On Metadata

Indexes play an important role in measuring comprehensive development and other fields.People construct multiple index systems for scientific management and effective monitoring in many industries,such as social economy,education,science,agriculture,defense and so on.With the development of big data technology,using data to do managements and decisions is getting more and more important.Researchers pay more attention to index system,and have to spend a lot of time on producing indexes,selecting indexes and other steps of constructing index system.At present,many departments have accumulated a large number of indexes for specific tasks after long-term research,and a large number of papers have been written from the research of indexes in various fields.Many indexes can also be extracted through automated extraction technology.All of indexes which have been accumulated can provide important reference for the construction of the new index system.However,due to the wide range and large number of indexes,it is difficult for experts to find relevant indexes quickly.Therefore,it is necessary to study how to find task-related indexes from the massive indexes and recommend them to experts.It is of great significance to construct a new index system.The article mainly uses deep learning technology to vectorize specific tasks and index metadata and then uses text matching technology to calculate the similarity between index vector and task vector.For the task,the most relevant index can be recommended to help experts to select indexes quickly.(1)A basic framework for index selection based on metadata is Designed.At present,there are few studies on the selection of indexes.The article firstly defines the index selection problem as the matching problem between indexes and tasks.According to the process of information representation,similarity calculation,and similarity ranking,a metadata-based approach is established.The index metadata that can be used for selection is designed to describe the index,and a task representation is designed to describe the task.(2)Two index selection methods are proposed for different application scenarios.When the task information of the index is sufficient,the fully connected neural network is used to dig out the matching relationship between the index information and the task information to realize the index selection;when the task information of the index is missing,the training model lacks effective data support.Firstly,the task information of the indexes can be marked by the field staff.The siamese network is combined with the matching relationship between the index information and the task information to realize the index selection.The index selection method has strong flexibility and practicability.(3)The index selection experiment analysis is carried out on the national defense science and technology index data set.The data set contains 2179 indexes.The experiment is to validate the index selection method based on the fully connected neural network and the index selection method based on the siamese network.The experimental results show that the first method can recall 72.6% indexes in the first 50 Indexes,and the second method can recall 72.7% indexes.In summary,the article focuses on the research of index selection methods.Based on index metadata,an index selection method based on fully connected neural networks and an index selection method based on siamese networks are proposed,which can solve the actual index selection process to some certain extent.The index selection method based on metadata have theoretical and practical significance for the development of index selection technology.