This invention discloses an intelligent filtering and completion method and
system based on a large
language model and hierarchical mapping, relating to the fields of
artificial intelligence and
semiconductor manufacturing execution and
data retrieval. The method includes initialization configuration, slot extraction and preprocessing, hierarchical mapping loading and candidate replacement, core filtering key
verification and device retrieval, filter combination determination and automatic completion,
missing data detection and interactive action determination, response construction and return, and
exception handling steps. The
system includes a front-end dialogue and interface, a large
language model service module,
a site-factory-region caching module, a device retrieval and
verification module, a strategy engine, and a
response generation module. This invention tightly integrates the
natural language understanding capabilities of a large
language model with back-end structured resource
verification, supporting multi-turn dialogue and fine-grained automatic completion, significantly reducing invalid queries, improving retrieval
hit rate and user experience, adapting to complex multi-site
semiconductor manufacturing scenarios, and featuring a
modular design with strong
scalability.