UDC 004.8
METAHEURISTIC APPROACH TO AUTOMATIC SYNTHESIS OF A GENERALIZED METRIC FOR PAIRWISE COMPARISONS OF INTELLECTUAL ACTIVITY PRODUCTS DURING CLUSTERING
K. A. Potapova, Lecturer, ERP Systems Department, Institute for Information Technologies, Federal State
Budget Educational Institution of Higher Education «MIREA – Russian Technological University»,
Moscow, Russia;
orcid.org/0009-0000-6573-8688, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
A. V. Gorchakov, PhD (in technical sciences), Associate Professor, ERP Systems Department, Institute
for Information Technologies, Federal State Budget Educational Institution of Higher Education «MIREA –
Russian Technological University», Moscow, Russia;
orcid.org/0000-0003-1977-8165, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The process of effective professional team formation requires taking into account many factors: from the
competencies of participants to their style of work and the compatibility of approaches they use to solving
problems. The aim of the work is the development of a metaheuristic approach to forming a generalized criterion
for pairwise comparisons of intellectual property products, which allows, by optimizing the weights of
metrics used together, to minimize the discrepancy between automatically obtained assessment and expert
assessment of pairwise similarities of scientific articles or software code. A comparison of four optimization
algorithms, such as genetic algorithm, particle swarm optimization algorithm, differential evolution algorithm,
and nuclear reaction optimization algorithm, was conducted based on their ability to find nearoptimal
solutions. For scientific articles, weight coefficients were optimized for cosine similarities between:
vectors which components correspond to the frequency of occurrence of words in texts (Term Frequency-
Inverse Document Frequency, TF-IDF); contextual vector-based text representations formed by BERT (Bidirectional
Encoder Representations from Transformers); vector-based text representations based on the Continuous
Bag of Words (CBOW) model, which is part of the Word2Vec architecture. For source code, weight
coefficients were optimized for structural metrics, such as the number of lines of code, cyclomatic complexity,
the number of functions, token-based metrics, and frequency histograms of syntax tree node types. During
the conducted computational experiment, we found out that the generalized metric with weighting coefficients
configured by particle swarm optimization algorithm improves the correlation with expert assessments
by 28.61 % for scientific articles (r = 0,8013) and by 20.11 % for code (r = 0,7604) when compared to the
best results obtained from individual metrics used separately. The proposed approach can be applied to
solving such practical problems as the automated creation of teams of specialists, the selection of reviewers
for scientific papers or software source code; the creation of student groups, and can be adapted for the
analysis of other types of structured data, such as technical documentation or patents.
Key words: metaheuristic approach, similarity metrics, particle swarm optimization, source code analysis,
scientific texts analysis, abstract syntax trees, automated team formation.