Intellectual Property India Publishes Patent Application for Topology-Adaptive Score Fusion With Temporal-Decay Graph Centrality For Anti-Money Laundering Risk Prioritization

MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202641087869 A) filed by Vellore Institute Of Technology on July 17, 2026, for Topology-Adaptive Score Fusion With Temporal-Decay Graph Centrality For Anti-Money Laundering Risk Prioritization.

Inventors include Varalakshmi M; Ranichandra C; Thilagavathi M; Ankit Subedi; Abhishek Kumar; Aditi Sharma; Pihu Gupta; and Bistrit Koirala.

The application for the patent was published on July 24, 2026, under issue no. 30/2026.

Abstract: ABSTRACT TOPOLOGY-ADAPTIVE SCORE FUSION WITH TEMPORAL-DECAY GRAPH CENTRALITY FOR ANTI-MONEY LAUNDERING RISK PRIORITIZATION A computer-implemented method and system for anti-money laundering risk prioritization receives, via a Transaction Data Input (110), transaction data representing a directed financial transaction graph having nodes representing accounts and edges representing transactions. The method computes a temporal-decay PageRank score for each node by performing a power iteration wherein edge weights are computed as a product of the transaction amount and an exponential temporal decay factor. The method applies burst-velocity amplification to edge weights for sender nodes and applies a directional strongly connected component penalty to nodes within strongly connected components using asymmetric dampening factors for collector nodes, distributor nodes, and balanced nodes. An Ego-Network Extractor (112) extracts a topology vector for each account from a 2-hop ego-network. A topology attention gate computes per-account fusion weights by applying multiplicative gating on an ensemble weight dimension. A fused risk score is generated by combining a machine learning score from a LightGBM Classifier (80), the temporal-decay PageRank score, and a rule-based score from a Symbolic Rule Engine (130) using the per-account fusion weights. A Degradation Controller (140) selects an execution path from a routing table based on component health status and a precision budget constraint. (FIG. 1)

Disclaimer: Curated by HT Syndication.