Re:Apple switches from Bing to Google for Siri web search on...
by tesseract on Tuesday September 26, 2017 @01:54AM
(#15336509)
For true long tail queries it seems to me that often (speaking only as a Google user, not an expert) the NLP really ends up getting in the way by causing the query to be reinterpreted as something more mainstream. But I agree that it's important for getting high quality results for a lot of common queries. And that quality and relevance scoring of results is also a hard problem.
In search industry, you can probably spend 1x time and resources to reach 90% Google's relevance score. However, to reach 95% or more, you probably need 100x in time and resources.
E.g., NLP in query understanding; Knowledge Graph in webpage understanding; Much larger corpus; Location awareness, etc.