<![CDATA[General Background Modern digital repositories face severe operational constraints due to the exponential growth of born-digital materials and digital collections. Specific Background Traditional data organization systems struggle to handle massive digital volumes, leading to metadata deficiencies, processing backlogs, and compromised asset accessibility. Knowledge Gap Existing digital preservation strategies focus primarily on isolated automation tasks, lacking an integrated, multi-phase framework that combines literature landscape tracking, automated catalog error correction, and semantic linked data expansion. Aims This study develops a comprehensive computational framework leveraging machine learning, natural language processing, and knowledge schemas to optimize data workflows in multilingual institutional libraries. Results Utilizing a calibrated Random Forest Classifier paired with custom character-level n-gram frequency segmentation algorithms, the system achieved macro ROC-AUC scores of 0.958257, effectively isolating substandard catalog records and correcting phonetic transliteration errors in Cyrillic indexing systems. Novelty This framework introduces a unique dual-layer pipeline that resolves the physical contextual fragmentation of archives while simultaneously rectifying multilingual transmission flaws without requiring external spell-checking dictionaries. Implications The integration of intelligent automation structures provides memory institutions with scalable, reproducible asset tracking mechanisms that enhance metadata discoverability while preserving the essential contextual validation role of professional archivists. Keywords: Computational Archiving, Machine Learning, Information Retrieval, Metadata Optimization, Digital Preservation Key Findings HighlightsAutomated Random Forest pipelines achieved a 0.958 macro ROC-AUC efficiency in identifying corrupted catalog metadata records. Character-level n-gram frequency mapping effectively isolates phonetic translation inconsistencies within multilingual indexing systems. Cognitive graph modeling successfully reconstructs structural provenance patterns lost during analog-to-digital asset transformation.]]>
