Entity-Driven Alerts Using Disambiguated Features

The patent segments the monitoring process into distinct modules: feature extraction module that identifies potential entities, disambiguation module that resolves ambiguity by analyzing context and relationships, scoring module that ranks results, and linking module that connects entities to documents. This segmentation allows broad coverage while maintaining precision through specialized processing at each stage.

The patent introduces intermediary components between keyword search and alert generation: a knowledge base that stores entity relationships and a disambiguation system that acts as a mediator to filter and validate entity references. These intermediaries enable broad monitoring coverage while ensuring alert precision by verifying entity-context matching before generating alerts.

The patent divides the complex disambiguation task into separate functional modules: feature extraction, disambiguation, scoring, and linking. Each module handles a specific aspect of the problem, making the overall system more manageable despite the increased precision requirements. This modular approach reduces system complexity by organizing complexity into discrete, reusable components.

The patent performs preliminary actions by pre-building a knowledge base of entity relationships and co-occurrence patterns before the actual monitoring begins. This pre-processing step stores commonly co-occurring features and entity associations, allowing the disambiguation module to quickly resolve ambiguities without performing complex analysis in real-time, thus managing system complexity while maintaining high precision.

The patent changes the parameters of alert generation by introducing a scoring mechanism that evaluates the relevance and significance of each entity mention. Instead of alerting on all mentions equally, the system adjusts parameters such as confidence thresholds and relevance weights to filter out low-value alerts. This maintains information completeness for significant events while reducing overall alert volume through parameter-based filtering.

The patent applies partial action by selectively generating alerts only for entity mentions that meet specific criteria (e.g., high confidence scores, novel associations, significant co-occurrences). Rather than alerting on every possible entity mention, the system performs partial processing that focuses computational resources on the most valuable alerts, reducing information overload while preserving completeness for important information.