KMAN Publication Inc. AI and Tech in Behavioral and Social Sciences In Press 2026 10 01 Enhancing BPMN Repository Consistency Using Hybrid Similarity Metrics and Scalable Retrieval Enhancing BPMN Repository Consistency Using Hybrid Similarity Metrics and Scalable Retrieval 1 11 EN Journal Article 2026 03 10 <p>Business Process Model and Notation (BPMN) has become a dominant standard for representing, communicating, and automating organizational workflows. Yet the growth of large process repositories has intensified inconsistencies caused by heterogeneous modeling conventions, duplicated fragments, inconsistent activity labels, and structurally similar models represented in divergent ways. These problems reduce process reuse, weaken repository governance, complicate integration, and increase maintenance costs. This study proposes an integrated framework for enhancing consistency in BPMN repositories by combining hybrid similarity assessment with repository-level recommendation mechanisms. The framework converts BPMN models into directed graphs and computes multilevel structural similarity using dynamic vector signatures inspired by BPMN-Sim. Semantic similarity is calculated from Sentence-BERT embeddings and cosine similarity, allowing conceptually equivalent labels to be detected even when lexical forms differ. The resulting hybrid similarity score is used by recommendation modules that propose standardized labels and reusable process elements during modeling. HDBSCAN clustering and HNSW indexing are incorporated to support scalable candidate retrieval. The approach was evaluated using public BPMN repositories and benchmark tasks. Results show that the proposed method outperforms structural-only, lexical, and semantic-only baselines. After label standardization, the method achieved Precision = 0.91, Recall = 0.90, and standard F1 = 0.90; after element standardization, it achieved Precision = 0.90, Recall = 0.88, and standard F1 = 0.89. The findings indicate that similarity assessment should be treated not only as a retrieval technique but also as a mechanism for repository standardization, reuse, and intelligent process modeling.</p> https://journals.kmanpub.com/index.php/aitechbesosci/article/download/5792/11353