Abstract
This article presents an original model for solving reference in natural language texts, which uses domain knowledge to improve accuracy. The novelty of the approach lies in the integration of machine learning methods based on neural networks with domain ontologies to resolve ambiguous references. Traditional reference decision models are often limited to using only linguistic features, which leads to errors in complex texts, especially in specialized fields such as medicine or law. The proposed model includes several key components: text preprocessing, selection of references, prediction of antecedents using language models and refinement of predictions based on knowledge of the subject area. In the course of the research, a methodology for the application of ontologies was developed, which improves the resolution of references in complex texts using contextual knowledge. The results of experiments conducted on corpora of texts from various subject areas showed a significant increase in accuracy compared to basic models that do not use knowledge of the field. The use of this model is especially effective for texts with a high degree of professional specificity, where lexical ambiguity and complex linguistic constructions can complicate the process of solving the reference.