Article Preview
Top1. Introduction
Modern TNs are complex technical objects that are characterized in terms of their structure, functionality, communication channel types and their capacity, interaction with other objects, users, etc. The structures of the networks and their status are constantly changing. As result, building TN models requires many resources both human and computational.
There are three main classes of problems that are solved by the operators of the telecommunications networks. These are search, analytical, and recommendation (forecast) problems. The first class of problems assumes that data about the TN which meets a given set of requirements is obtained. To solve this class of problems known links between the nodes of the networks are used. For solving the second class of problems it is necessary to restore links from the data about the networks that is provided by various data sources. Solution of the third class of problems requires identification of new links based on existing ones.
Nowadays for real life TN problems solving a number of graph models are used. Each graph model contains data of one type (Krinkin et al., 2020). But now TN operators are interested in solving the tasks using general models where links between data elements of different types are defined. These models should also contain operational data about the networks. The operational data include user actions, service invocations, service performance statistics, incidents and other events that are monitored in the networks. To meet the modern requirements of TN operators, new KGs that contain both static and operational data about the networks were proposed (Krinkin et al., 2020; Yoshinov et al., 2020). In (Osipov, Lushnov, Stankova et al, 2017; Osipov et al., 2019) new inductive synthesis methods for building TN models were considered. These methods are capable to transform and link the data of different types and provide it in the form required by the users. Using these models users can solve search and partly analytical tasks. They don't allow solve a considerable number of analytical and recommendation tasks that require forecasting.
One of the common forecast tasks solved by the operators of telecommunications networks is analysis of trends in behavior of millions of users in the context of provided services, used applications and users interests in content types (Krinkin et al., 2020). For solving this type of tasks, it is necessary to determine the possible set of the network states in different conditions. It means doing the trend assumption and check it for a certain point in time. The same tasks are solved for determining trends of network performance. The problems considered above can be solved using deductive synthesis of the TN models. One of the main problems of using deductive synthesis methods for solving analytical and recommendation tasks is their high computational complexity. A new deductive synthesis method of hierarchical TN models with low computational complexity for solving forecast (prediction) problem is introduced and discussed in the article.
The new suggested method of deductive synthesis with lower computational complexity is designed for objects with hierarchical structure (Yoshinov et al., 2020). Despite the fact that this approach is quite promising for building models of the networks, it has not yet been given proper attention in either theory or practice in application to knowledge graphs synthesis. The reason is that for a long time it was possible to solve real-life problems by creating a large number of highly specialized solutions. However, this approach has nearly run its course due to the ever-increasing complexity of objects and problems being solved and frequent changes in the requirements of end users to the results. The article discusses a new deductive synthesis method that allows synthesize hierarchical models of telecommunications networks that are described in the form of hierarchical knowledge graphs (Sarrafzadeh & Lank, 2017). Suggested methods provide speedup up to 50% depending on the structure of the model against non-hierarchical approaches.