Watchlist Screening Using Two-Phase Candidate Filtering

The patent segments the matching process into two distinct phases: (1) an initial candidate identification phase using a first comparison technique that transforms attribute values into sequences of textual segments and compares them efficiently, and (2) a refinement phase using a second comparison technique that performs more accurate but computationally intensive analysis. This segmentation allows the system to balance speed and accuracy, reducing false positives by applying the more accurate second technique only to promising candidates rather than all possible pairs.

The patent implements dynamic adaptability by allowing the system to learn from analyst feedback and validate decisions. The machine learning model is trained on validated match data, enabling the system to adapt its matching criteria over time. This dynamic learning process continuously improves match accuracy and reduces false positives by adjusting to evolving criminal strategies and patterns.

The patent divides the comparison process into two sequential stages with different objectives. The first comparison technique uses transformation of attribute values into textual segment sequences for efficient initial filtering, achieving high productivity. The second comparison technique applies more rigorous analysis to the narrowed candidate set, achieving high accuracy. This segmentation resolves the contradiction by applying different methodologies appropriate to each stage of the filtering process.

The patent applies the principle of partial action by using the computationally intensive second comparison technique only on a subset of candidates identified by the first technique, rather than applying it to all possible pairs. This partial application of the more accurate method achieves high overall accuracy while maintaining efficiency, as the expensive operation is performed only where necessary.

The patent performs preliminary filtering using the two-phase comparison technique to identify and narrow down candidate matches before they reach manual verification. By applying efficient initial comparison followed by more accurate refinement, the system pre-processes the data to present only the most promising candidates to analysts. This preliminary action significantly reduces the volume of work requiring manual verification while maintaining high accuracy.

The patent implements self-service through automated machine learning validation that learns from analyst feedback. The system automatically trains its own models using validated match data, enabling it to improve its performance without requiring continuous manual intervention. This self-learning capability reduces the long-term burden of manual verification while maintaining or improving accuracy over time.

The patent implements dynamic adaptability through machine learning models that continuously learn from validated match data and evolving patterns. The system automatically adjusts its matching criteria based on learned patterns, enabling it to adapt to new criminal strategies without requiring manual reconfiguration. This dynamic learning process maintains ease of operation while achieving high adaptability.

The patent incorporates feedback loops where analyst validations of match decisions are fed back into the machine learning model for retraining. This feedback mechanism allows the system to learn from real-world outcomes and continuously improve its matching criteria. The feedback-driven learning process enables the system to adapt to evolving criminal strategies while maintaining operational simplicity, as the complexity of adaptation is handled automatically by the learning algorithm.