A finite-state transducer (FST) is a compact lookup structure for matching many term sequences quickly. In semantic search, it supports entity extraction from incoming query text when the known-entity list may contain millions of phrases. section 7.4.1
The search pipeline stores surface forms, canonical forms, entity types, spelling variations, and popularity values in an entities collection. Solr can build an FST-backed lookup index from that collection, then tag matching phrases in milliseconds. section 7.4.1
Entity extraction from a query
For “top kimchi near charlotte,” the extractor tags “top,” “near,” and “charlotte.” Popularity can select the location interpretation of “near,” producing a typed query tree containing a popularity function, the keyword “kimchi,” a location-distance function, and Charlotte.
