Aerospace Instrument-Making Annotation << Back Validation of the Mode Applicability of TF-IDF and Context Blurring in Matching Weakly Structured Incident Texts Rodionov D.G., Konnikov E.A., Golikov G.I. This article focuses on improving the analytical suitability of weakly structured technical incident texts for risk management and management decision support tasks. The authors consider the key limitation of classical TFIDF—its sensitivity to wording variability and inability to restore semantic connections between terms, which leads to sparse representations and fragmentary thematic structures when comparing incident reports. A TF-IDF method with contextual blurring is proposed, which enriches the vector representation of documents by diffusely distributing weights across a graph of term associations based on their joint occurrence, thereby reducing the effect of false orthogonality. Testing on a corpus of specialized reports on incidents in the nuclear power industry showed that, with a moderate compromise on formal geometric clustering metrics, a significant increase in semantic connectivity and interpretability of thematic structures is achieved, as well as more stable identification of local analogies in modes of weak lexical overlap. The proposed approach provides a transition from statistical grouping to the extraction of semantically coherent patterns of incidents relevant to control regulations and subsequent managerial interpretation. Keywords: thematic modeling, TF-IDF, contextual blurring, semantic regularization, semantic proximity, co-occurrence graph, text vectorization, weakly structured incident texts, thematic coherence, natural language processing (NLP), regime applicability, risk management, management decision support. DOI: 10.25791/aviakosmos.4.2026.1552 Pp. 47-60.