Automatic Routing of Official Documents Entering the Organization’s Executive System Based on NLP Transformer Models and Semantic Segmentation
Year: 2026
Type: conference-paper
Abstract: The rapid growth of document flow in organizational execution systems creates significant challenges in the timely and accurate routing of official documents to responsible departments. A single document often contains multiple thematic segments related to different structural units, which complicates manual processing and reduces operational efficiency. This study proposes an artificial intelligence-based algorithm for automatic classification and routing of official documents in organizational execution systems. The proposed approach combines semantic segmentation of documents with transformer-based text embeddings to improve classification accuracy and contextual understanding. Unlike classical topic modeling methods such as LDA, which require large corpora and demonstrate limited generalization capabilities, the proposed method leverages pre-trained transformer models to achieve high performance even with a moderate-sized dataset. The experimental dataset consisted of 200 official documents, which were segmented into more than 1,200 semantic fragments. Three models were evaluated: LDA (baseline), BERT-based classification, and the proposed segmentationenhanced BERT model. Performance was assessed using Precision, Recall, and F1-score metrics. Experimental results demonstrate that the proposed approach significantly outperforms classical methods, achieving an F1-score of 0.91 compared to 0.50 for LDA and 0.84 for standard BERT classification. The findings confirm that semantic segmentation combined with transformer models provides an effective and scalable solution for intelligent document routing in execution management systems. The proposed approach can be applied in governmental institutions, higher education institutions, and large organizations to improve document processing efficiency and execution discipline.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202292587