Modeling of an Intelligent Algorithm for Ranking Authorized Institutions in Response to Requests

The presented work explores the application of semantic matching methods for vector representations of text to create an intelligent service that can rank authorized entities in the task of preparing responses to requests. The proposed system is modeled using the example of identifying performers for preparing responses to citizens' appeals to government agencies. The system is designed to not only identify relevant performers but also provide an explainable connection between the content of the appeal and the authority of the government agency, making it an effective tool for professionals. A hybrid architecture is considered, which includes a pre-trained neural network model based on the SBERT architecture. The results demonstrate a significant improvement in ranking quality: the final model achieves MRR=0,996 on the training set and 0,739 on the test set, which confirms the applicability of the developed approach for automating the process of processing citizens' appeals.

Keywords

explainable artificial intelligence; semantic matching; ranking models; natural language processing.

References

1. Edverest. Obrabotka estestvennogo yazyka i neyronnye seti [Natural Language Processing and Neural Networks]. Available at: https://edverest.ru/unit/obrabotka-estestvennogo-yazyka-i-nejronnye-seti/ (accessed on: 31.01.2026) (in Russian)
2. Shalagin N.D. [Review of Semantic Search Algorithms for Text Documents]. Mezhdunarodnyy zhurnal otkrytykh informatsionnykh tekhnologiy [International Journal of Open Information Technologies], 2024, vol. 12, no. 9, pp. 11-20. Available at: https://cyberleninka.ru/article/n/obzor-algoritmov-semanticheskogo-poiska-po-tekstovym-dokumentam (accessed on: 15.02.2026) (in Russian)
3. Kitov V.V. Latentnyy semanticheskiy analiz [Latent Semantic Analysis]. Available at: https://deepmachinelearning.ru/docs/Neural-networks/Embeddings/Latent-semantic-analysis (accessed on: 11.01.2026) (in Russian)
4. Nefedova A.A. [Model' II dlya ob"yasnimoy marshrutizatsii obrashcheniy grazhdan v organy vlasti] [AI Model for Explainable Routing of Citizens' Appeals to Government Bodies]. Iskusstvennyy intellekt: nauchnye dostizheniya i prikladnye zadachi: Materialy II Vserossiyskoy nauchno-prakticheskoy konferentsii s mezhdunarodnym uchastiem [Artificial Intelligence: Scientific Achievements and Applied Problems: Proceedings of the II All-Russian Scientific and Practical Conference with International Participation]. Khanty-Mansiysk, Yugorskii Gosudarstvennyi Universitet, 2026, p. 103. (in Russian)
5. Reimers N., Gurevych I. Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks. arXiv. Available at: https://arxiv.org/abs/1908.10084 (accessed on: 02.02.2026)
6. Feng F., Yang Y., Language-Agnostic BERT Sentence Embedding. arXiv. Available at: https://arxiv.org/abs/2007.01852 (accessed on: 30.01.2026)
7. ShAD. Available at: https://education.yandex.ru/handbook/ml (accessed on: 11.11.2025) (in Russian)
8. SentenceTransformers Documentation. Available at: https://sbert.net/ (accessed on: 21.10.2025)