Named Entity Recognition and Ticker Linking

Before a model can react to "Apple announces new product," it first has to recognize that "Apple" is a company name, not a fruit, and that "Meta" in a headline is Meta Platforms rather than an ordinary English word. Named entity recognition (NER) is the step that tags spans of text as belonging to categories like company, person, or location; ticker linking (or entity linking) is the follow-up step that maps a tagged company mention to the specific security it refers to.

Ticker linking has to resolve several kinds of ambiguity: common words that are also company names ("Meta," "Delta"), multiple listed share classes for one company (Alphabet's GOOGL and GOOGT), private subsidiaries mentioned by their parent's public ticker, and companies that share nearly identical names across different exchanges.

Worked example. A news headline reads: "Delta reports record quarterly revenue after fare hikes." An NER model tags "Delta" as an organization. The ticker-linking step then must decide, using surrounding context like "fare hikes," that this refers to Delta Air Lines (DAL) rather than Delta Electronics or a company simply named Delta Corp — a decision that would fail if the linker only matched on the bare string "Delta."

Errors at this stage propagate downstream: a sentiment or event-detection pipeline that mislinks a mention will attach the wrong signal to the wrong ticker, so ticker-linking accuracy is often the actual bottleneck in a news-driven trading system, not the sentiment model itself.