MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202621048855 A) filed by Beyondata Solutions Private Limited on April 16, 2026, for System And Method For Machine-Learning Based Cross-Page Identifier Consistency And Normalization.
Inventors include Nishant Singh Tomar; and Dipesh Prajapati.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: The present disclosure relates to a system (114) and method for cross-page identifier consistency verification and canonical normalization across multi-page documents. The system (114) receives extracted textual tokens with positional metadata and identifies candidate identifiers using rule-based templates and sequence-labeling models. An aggregation module (212) constructs a global representation of identifier occurrences across pages. An equivalence-scoring module (214) computes identifier-equivalence scores using lexical-normalization similarity, contextual-embedding similarity derived from a domain-specific small language model (120), and spatial priors. An anomaly-analysis module (216) detects typographic or structural inconsistencies, and a normalization module (218) generates canonical identifier forms. A consistency-resolution module (220) refines corrections using global conflict analysis. An output module (222) produces the normalized identifiers with confidence scores and justification metadata linking each canonical form to page-level evidence. The system (114) improves identifier accuracy, consistency, and auditability across large document sets. FIG. 1
Disclaimer: Curated by HT Syndication.
