Snippet extraction

Snippet extraction is the process by which a short passage is selected from a document to be displayed as an answer or as a preview of the source. The selected sentence becomes your calling card – the user often reads it instead of the whole page and decides whether to click based on it. The system favours a sentence that answers directly, is understandable without the surrounding text, and contains a specific figure. Sentences that begin with a pronoun or a reference to the previous paragraph fail, as do sentences with a condition stated only at the end, and marketing phrasing without information. In practice, it is worth placing one complete sentence under each heading that answers the question in the heading, followed only then by the elaboration. Such a sentence holds up even when pulled out of context.

See also: Passage indexing, Definition Sentence at the Beginning, Content chunking.