What is the future of semantic search?

Matt's answer:

CUTTS: We have a kind of a delicate question from Magico in London, UK. Magico asks “What is the future of Semantic search?” That’s a really hard question not only because it’s difficult to predict the future, but also because Semantic search is kind of a overloaded term and different people think about it in different ways. So some people think about Semantic search in a way in which website creators or publishers will be able to specify relationships. So traffic lights are red, you know, rain falls down. That’s sort of a conventional way of thinking about Semantic search. There is even this notion about database of facts where for 10 years everybody can build common sense facts and sort of write different rules. That has had a lot of promise, but there’s the danger that it’s kind of brittle. And so, you’ll sometimes see a Peter Norvig has done a very good job where he’s talked about Semantic search in terms of statistical data. So rather than writing a dictionary where you say here are all the different ways you could translate the word “Set” which has a lot of different definitions. Instead, Google might look at all the different ways were you can find out, okay, here’s someone writing in English, here’s someone writing in French. And you can sort of statistically take the content of the web and sort of learn in aggregate, oh, these words correlate a lot. It turns out this word might be a good synonym in this language for this word. So there has been this long writing discussion about whether Semantic search will be people specifying almost like Meta tags, you know, like the actual data or whether the machines will learn it probabilistically or statistically. Google tends to lean toward the idea that, you can have relatively simple algorithms. And if you have a lot more data like a hundred times or 10 times as much more data, you can still qualitatively do really, really interesting things. So historically, we lean a little more towards the probabilistic or statistical model way of looking at things. As far as the future of search at Google I mean don’t be at all surprised semantically if we try to do more query understanding. So if you type in, you know, “Car battery died,” and we find–we might rewrite that query a little bit and say “Automobile battery died.” And then also document understanding so you can have a same sort of words used in a document and we can say, “Oh, its okay if we can match automobile and car or maybe a specific model of car?” And so we’ll keep working on trying to figure out user intent; what users are asking, not just the keywords that they’re using, and also document understanding. So if a document happens to be talking about 24-inch dishwashers and doesn’t specified that it’s 24 inches, maybe we can infer that someway or maybe we can find out inch and, you know, the little quotation marks are the same. So we’ll keep working on all of those sort of things. If you were, as a webmaster, trying to make it more likely for your document to show up, try not to make the search engines do a ton of extra work. Think about what are the words users are going to type whenever they try to find the information on your page and include those words but not in a spamming way, not in a keyword stuff way at the bottom. Just think about some of the ways that people will ask and just fit those into the copy of your website in a natural way where you’re not stuffing and you’re not trying to like squanch a whole bunch of different stuff in there just like the different ways that people could talk about it. If you read it out loud and it sounds pretty natural and it sounds conversational, but you still found a way to get three or four synonyms in or different ways that somebody could do that query, that’s probably going to be a pretty good page as far as all the different ways that users will ask that question.

by Matt Cutts - Google's Head of Search Quality Team

Original video: