Neural Network Query Tagging for Polysemous Term Disambiguation

The patent segments the query understanding task into multiple specialized neural networks, each responsible for specific entity types (e.g., one network for company entities, another for software entities). This segmentation allows each network to specialize in recognizing specific meanings of terms, thereby resolving the ambiguity of polysemous terms like 'Acme' while maintaining high precision in entity recognition.

The patent introduces an intermediary layer of entity type predictions between the basic named entity recognition and the final search query understanding. This intermediary layer provides additional contextual information about the specific type of entity being referenced, enabling the system to distinguish between different meanings of the same term by predicting whether 'Acme' refers to a company, software product, or other entity type.

The patent divides the query processing system into multiple specialized neural networks that operate in parallel, each trained on specific entity types. This segmentation maintains processing efficiency through parallel computation while improving result relevance by having each network specialize in recognizing specific entity meanings, thereby reliably distinguishing between different interpretations of ambiguous terms.

The patent changes the parameters of the recognition system by training separate neural networks with different parameter sets optimized for specific entity types. Each network learns distinct patterns and features relevant to its specialized entity type, enabling the system to maintain high processing efficiency while achieving reliable differentiation between multiple meanings of the same term through parameter specialization.