Every Tuesday and Thursday morning, Google’s search engineers meet to talk about the algorithm that forms the heart of the company’s multibillion-dollar business. They discuss the tiny tweaks, of which there are about 600 each year, that keep Google in its place as the world’s favourite search engine. These are the people who guard the company’s status as, in the words of ranking team head Amit Singhal, ”the biggest kingmaker on this earth”.With that unique position, however, comes responsibilities – when Google makes small changes, users’ behaviour can alter enormously. The fortunes of whole companies and countless careers live and die by their rankings in search results.”We deal with those responsibilities by having very concrete principles,” says Singhal. ”All rankings are decided algorithmically, and the focus is on user benefit, not advertiser or commercial benefit. We ask ourselves, ‘Can a random company that does not want to be part of any Google system be harmed by a change we’re proposing?’ If they are, we won’t do it.”Last week Google said it was finally implementing the new way of looking at the web that it had announced last year, called Caffeine. The idea is that the company crawls the web constantly producing an index in real-time – which means that users’ search queries are always being directed towards a new version of the index, rather than, as previously, a version that could be up to two weeks old.The change, imperceptible to Google’s end users, should lay the groundwork for the site’s continued place at the forefront of search technology, handling upwards of 400 million inquiries a day.Udi Manber is the company’s vice-president of engineering, and he has been studying search for more than 20 years. ”People’s expectations have grown,” he says. ”Ten years ago, when you looked for something and found it, you’d be really impressed. Now, when you don’t immediately find exactly what you want, you think something’s broken.” He points out that the equivalent process took aircrafts a century.Now, however, results are almost a given – from Microsoft’s Bing to Ask出售老域名, the race has evolved. In part, it is about producing real-time results from social networks such as Twitter, but the challenge, too, is to work out what people want to search for next.Manber estimates each internet search requires the same amount of computing power as was available globally in the ’50s, and queries are getting more complicated.The problem Google faces is not simply finding a mathematical model for language, however. ”When it comes to human language understanding, we are still at the toddler stage,” says Singhal. ”But over the next 10 years we will attain the level of an eight- or nine-year-old. We’ll be able to perfect experiences we don’t fully trust today.”That means speech recognition that works, and putting databases together so that they play nicely. The results, Singhal says, ”sound like science fiction”.”A search, say, for chocolate milk on your internet-enabled phone,” he says, ”would produce directions to the nearest store that had it in stock – any information, anywhere you are, comes to you before you need it. The world is not web pages – it is entities and things.”Indeed, it seems what excites Singhal today is this integrated experience. ”It will be important for machines to have a good grasp of what you really want. So currently, when I am logged in Google knows what sports I’m into. In a privacy-preserving manner, devices should be able to know a lot about me, because that brings a good experience.”The issue of privacy, however, is crucial, and after controversial problems with Google’s Facebook rival, Buzz, the company is keen to be seen to be transparently honest.With that ambition, says Manber, Google has made copies of the web for its own use. ”If an engineer has an idea, they can run their own experiment from their own PC.” He compares it to Boeing’s engineers being able to build their own aircraft and fly it all over the world. ”Which actually they can do now because you can do anything on a computer.”Telegraph, London
购买老域名