Building a Data Store with the Dynamic Structure

Abstract

This article presents the analysis of approaches to data warehouse construction based on relational and NoSQL solutions and lists the limitations of the relational approach to data mining. The contradiction between data presentation in the real subject domain and the model of data presentation in the relational and NoSQL approaches is revealed. The revealed contradiction is related to the temporality of the values of individual data attributes, the variability of the composition of these attributes, and structure of connections between them. A new logical model of the data warehouse with dynamic structure is proposed. The model is based on the concept of the object as a container for properties storage. Each property of the object includes the property name and two property values—without reference and with reference, that are relevant at a given time. The reference property value points to an object whose name is interpreted as the value of the property at a given time. A formal description of the model with allocation of the necessary functionality to manipulate objects and their properties (selectors, predicates, constructors) is given and the necessary control structures are introduced. Substantiation of the proposed model, called an OP-model is given on the basis of compliance with the logical ER data model. It is proved that any ER data model can be implemented in the OP-model. At the same time, the advantages of the OP-model are indicated, they are associated with the possibility of changing connections between entities due to changes in the reference value at a particular time. The potential for scalability of data warehouse due to the unique identification of each object is noted.