MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096146 A) filed by St. Peters Engineering College on August 08, 2026, for A Graph Neural Network-Based Intelligent System For Dynamic Relational Data Modeling And Prediction.
Inventors include Mrs. Muppalaneni Subhashini, Assistant Professor In Department Of Cseaiml, St Peters Engineering College, Opposite Ts Forest Academy; Mrs. P. Rajitha, Assistant Professor In Department Of Cseaiml, St Peters Engineering College, Opposite Ts Forest Academy, Kompally Road, Dullapally, Maisammaguda, Medchal, Hyderabad, Telangana; Mrs. Arukonda Srujana, Assistant Professor In Department Of; Mrs. G. Harikeerthana, Assistant Professor In Department Of Cse, St; Mrs. Jagadeeswari Tanukonda, Assistant Professor In Department Of; and Mrs. M Karuna Sudha, Assistant Professor In Department Of Mba, St Peters Engineering College, Opposite Ts Forest Academy Kompally Road, Dullapally, Maisammaguda, Medchal, Hyderabad, Telangana.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The present invention relates to a Graph Neural Network (GNN)-based intelligent system for dynamic relational data modeling and prediction, designed to analyze complex, interconnected, and continuously evolving datasets. The system addresses limitations of conventional machine learning approaches that treat data primarily as independent feature vectors and therefore may not adequately capture relationships, structural dependencies, and temporal changes among entities. The proposed system represents relational information as a dynamic graph comprising nodes, edges, node attributes, edge attributes, and temporal information. A data acquisition and preprocessing module collects and transforms heterogeneous relational data into a graph-compatible representation. A dynamic graph construction module continuously updates the graph in response to newly identified entities, relationships, interactions, and attributes changes. A GNN processing module performs message passing and neighborhood aggregation to learn structural and contextual representations from connected and multi-hop entities. A temporal relationship learning module incorporates historical and evolving interaction patterns into the learned representations. The resulting embeddings are provided to a prediction module configured to perform one or more tasks including node classification, link prediction, relationship forecasting, anomaly detection, recommendation, and future-state prediction. A model update mechanism continuously incorporates newly observed information and prediction outcomes to adapt the learning process to changing graph conditions. The system thereby provides relationship-aware, adaptive, and scalable predictive intelligence for dynamic relational environments. The invention is applicable to social networks, cybersecurity, financial transactions, healthcare, recommendation systems, transportation, communication networks, and Internet-of-Things applications.
Disclaimer: Curated by HT Syndication.
