COMPARATIVE ANALYSIS OF REVIEWER ASSIGNMENT ALGORITHMS FOR PHD-THESES COMMITTEES

In Ukraine, PhD theses are defended in one-time committees, which are formed manually. This can take a lot of time, and at the same time, the expertise of the selected members can be questionable, not to mention the corruption risks. Therefore, the automation of the process of assigning committees is important. In this work, a comparative analysis of algorithms for assigning reviewers based on assignment quality criteria and resource costs for finding members of the one-time committee is carried out. The subject of a thesis and the subject of the articles of potential committee members are represented by categorical distributions over research specialties from the ANZSRC-2020 research classification system. As a similarity criterion, improved Chekanovsky metric is used, which additionally takes into account the kinship of different research specialties using Jaccard index. The comparison of the efficiency of the algorithms for the assignment of committees is carried out experimentally on a dataset of 67 PhD theses. The NAQA information system was used to create the dataset. Experiments were conducted for the following optimization algorithms: brute force; brute force on a truncated set of candidates (only candidates with a certain minimum level of similarity are considered); greedy algorithm; greedy algorithm with elitism; isolated assignment (the most similar candidates are considered optimal). Thanks to optimization, it was possible to improve the composition of onetime committees by an average of 13-34%, depending on the algorithm used. This improvement occurred with a small candidate pool. An increase in the number of candidates will lead to a greater effect of applying the proposed algorithms for assigning opponents. As a result of the experiments, it was established that a good balance between the quality of assignment and resource consumption is achieved by a greedy algorithm without elitism and a brute force on a truncated set of candidates. At the same time, according to any algorithm, the newly created one-time committees are not worse, but usually better, than the one-time committees offered by the institution