Volume 4(53)
Bibliographic reference: Kalimoldayev M. N., Mazakova A. T., Yashchenko R. V., Mazakov T. Zh., Abdildayeva A. A. Electronic database design for zoological collection of the republic of Kazakhstan //journal “Problems of informatics”. 2021, № 4. P.40-54. DOI: 10.24412/2073-0667-2021-4-40-54
Novosibirsk State University, 630090, Novosibirsk, Russian Federation
ACTIVE KNOWLEDGE BASE PROTOTYPE ON THE BASIS OF COMPUTATIONAL MODELS
DOI: 10.24412/2073-0667-2021-4-55-66
The ever-growing volume of knowledge, for example, in the programming eld, requires a person to increase the speed of knowledge processing, to speed up knowledge mastering, and to use it more e‑ciently. One possible solution is to automate the process of knowledge application. But in many spheres of activity today humanity is accumulating knowledge in an informal way by using linguistic systems. Mastering the reading, understanding and correct use of the knowledge presented in this form, requires a long learning curve. Thus, knowledge accumulated in this form cannot be applied directly and automatically. Moreover, if the person who has mastered it does not use it very often, over time it will begin to be forgotten and subsequently can be completely lost.
In this work, the application of knowledge is considered as creation of a computer program that
utilizes it, and the automation of the knowledge application is understood as the automatic synthesis of such programs.
To solve mentioned problems e‑ciently the system must not only store the knowledge itself, but
also save the functional connections between individual concepts. In addition, it is necessary that the system, according to the task specication, is able to automatically construct a software application that solves it.
The system that allows one to automatically apply knowledge is called the Active Knowledge Base. This system is based on the theory of structural synthesis of programs. Knowledge in such a system is represented in the form of computational models, i.e. bipartite graphs in which the vertices in one set correspond to operations, and in the other - to variables. The edges of the graph determine whether a variable is the result of an operation or its input. Variables in such models correspond to some values of the described subject area, and operations are associated with certain program modules at the execution stage. The input and output parameters of program modules are associated with the variables of the computational model.
One of the key steps in constructing a software application is building a computation plan based
on the specification of the problem. A computation plan is a partially ordered set of operations, where the order relation is consistent with information dependencies and is defined as
"to compute F2, you need to compute F1\.
Within this prototype, knowledge in the form of computational models is stored in a special Sqlite database, and the system itself is divided into 3 parts. The rst part is a subsystem for storing computational models. It adds new computational models to the database and reads them from it.
The second part is a subsystem for constructing computation plans, which is engaged in building
computation plans according to the specification of the problem. The third part is an execution
subsystem. The computation plan is not a complete program that could be executed by the user's
OS. To execute it, one needs a system that interprets the plan. In this work it's done by the execution subsystem.
Computational models are described in special les, where each le is a description of one entity, for example, a variable of a computational model or an operation.
To construct a computation plan, a problem specication should be described rst. It consists of
the name of the computational model, a set of input variables, and a set of output variables. When the specication is completed, the le describing it is fed to the subsystem for constructing computationplans via command line arguments. Having received it, the system launches the construction algorithm, its result is a le with a computation plan.
The general operating principle of the execution subsystem is similar to that of the interpreter.
Operations are executed sequentially, as the variables on which they depend are ready. The execution subsystem supports the initialization of the input variables of the computation plan with the initial values passed to it; for this, it needs to create a special le with initialization parameters and send it to the execution subsystem via command line.
The process of operations execution continues until one of the conditions is met: all operations from the computation plan have already been calculated, or all output variables of the computation plan have been calculated. If in the course of calculations all operations from the calculation plan have been executed, and not all of the output variables have been calculated, then the calculations are completed with a corresponding warning message sent to the user. Otherwise, when the variables have already been calculated, and there are still operations in the computation plan, then the plan execution ends normally, the output variables are saved in accordance with their types, and the user is informed about the result of the computations through the standard output stream.
Results of the development and implementation of the system called Active Knowledge Base are
presented in the paper.
Key words: Active knowledge, Program synthesis, Active knowledge base, Structural synthesis of programs, Automatic programs construction, Knowledge base, Knowledge storage, Computational models.
Bibliographic reference: Artiukhov A. Active Knowledge Base prototype on the basis of computational models //journal “Problems of informatics”. 2021, № 4. P.55-66. DOI: 10.24412/2073-0667-2021-4-55-66
S. V. Bredikhin, V. M. Lyapunov, N. G. Scherbakova
Institute of Computational Mathematics and Mathematical Geophysics SB RAS, 630090, Novosibirsk, Russia
RANKING AUTHORS OF THE WEIGHTED CO-AUTHORSHIP NETWORK: ANALYSIS OF DB REPEC DATA
DOI: 10.24412/2073-0667-2021-4-67-83
This work was carried out under state contract with ICMMG SB RAS (0251-2021-0005).
In the previous paper [12] we investigated the co-authorship network N represented by an unweighted graph: nodes correspond to authors, and two authors are considered connected if they are coauthors of at least one publication. Basic network properties are: existence of the giant component (includes 90% of authors), small worldness [24] and a power-law tting of the distribution of coauthors. In this paper we focus on centrality measures in order to identify key authors on the base of the weighted co-authorship network. Using co-authorship data from the distributed database RePEc [13] we construct two weighted networks that dier in the way of computing edge weights.
Let P (|P| = l) be the set of publications and assume that each publication in P has at least
two authors. Let V (|V | = n) be the set of authors of these publications. We consider two schemes For assigning weights to edge when constructing networks. When using the “complete” weighting scheme the strength of the collaborative tie (the edge weight) between two aytors is set equal to the number of joint papers. AND when using the “fractional” weighting scheme the edge weight between two authors depend not only on the number of coauthored papers, but also on the number of other coauthors of these papers [7].
The raw data processing procedure is presented in [12], as a result the number of authors |V| = 32 434 and the number of coauthored publications |P| = 91 113. For each of the network
(unweighted and weighted according two schemes) four measures of centrality such as degree, closeness, betweenness and eigenvector have been calculated and the tables (tabs. 24) containing the names of the authors with the highest ranks are provided. It should be noted that these authors have high h-index values (according to Google Scholar search engine or IDEAS ranking system [25] based on all publications of the authors).
In order to study the dependence of author ranks on the method of calculating the contributions of authors to publications we calculated Pearson's correlation coefficients and Spearman's rank correlation coefficients for the same centrality measures for the networks under consideration. It was shown that regardless of how the edge weights are calculated the same centrality measures have signicant correlation with each other. The most signicant correlation according to both coefficients is fixed for the betweenness centrality, the least - for the eigenvector centrality, which determines the prestige of the network actor.
To illustrate the studied ways of calculating edge weights and the dependence of node ranks on the method and a node location, we considered the 12-node component of N and applied four centrality measures to its weighted representations. We see that the ranks of authors differ depending on the method of edge weights calculating. On the base of node ranks we calculated node weights and presented new ranks of authors (tab. 10) within any component representation and centrality measure used. It is noted that the high ranked authors are the influential persons with a large number of citations.
The purpose of further research is to identify the relationship between key authors and the number of citations of coauthored publications. The question of interest is whether collaborative publications receive more citations than single author publications.
Key words: bibliometry, co-authorship network, centrality measures, key authors.
Bibliographic reference: Bredikhin S. V., Lyapunov V. M., Scherbakova N. G. Ranking authors of the weighted coauthorship network: Analysis of DB RePEc data //journal “Problems of informatics”. 2021, № 4. P.67-84. DOI: 10.24412/2073-0667-2021-4-67-84
English
Русский