Page Zero: A Deep Dive

At Bing, our central mission is to help you spend less time searching and more time getting things done. To that end, we are in a constant state of experimentation looking for new ways to close the gap between the search you enter and the task you’re trying to get done. Whether it’s researching a vacation destination or buying a new watch, our goal is to figure out what you’re trying to do and help you get it done.

As we’ve all experienced, search is often a game of trial and error. For many searches, you have to try different variations of your query before you even get the right set of results. Then you’re tirelessly clicking on link after link to find the right web site that might have what you’re looking for. And after you’ve found the web site, it takes more time and steps to take an action like booking a plane ticket or making dinner reservations.

What if we could better understand and anticipate what you’re searching for? What if we could bring that information to you before you have finished typing? What if you could experience this in the blink of an eye?

Introducing Page Zero

Now you can, with Page Zero, which we introduced earlier this week. In the same 400 milliseconds it takes the average person to blink, Bing can suggest the best completion for a query and showcase rich information about the topic directly in the search box, without ever going to a results page. Harnessing Satori our underlying technology that lets Bing understand more about people, places and things, Page Zero introduces a faster way to search.

To bring this to life, we examined user interaction models, touch friendly interface design, deep understanding of user intent combined with Satori our underlying technology that lets Bing understand more about people, places and things.

Let’s watch Page Zero in action:

Entities and sub intents

In this case, someone is searching for “San Francisco Giants.” Here Bing provides a brief description of the baseball team, their logo and a list of sub-intents including “News”, “Images” and “Tickets”, “Scores” which are the most frequently searched for intents associated with the search phrase: “San Francisco Giants”.

image

Let’s see another example where someone searches for “Michelle Pfeiffer” and starts typing ‘michelle pf’. By harnessing Satori, Bing understands that Michelle Pfeiffer is an actress and therefore suggests highly related intents such as “Video”, “Biography” and “Filmography” which represent the most commonly searched tasks for queries related to movies and artists. Combining this with advanced machine learning techniques to determine the relationship between an incomplete query and the possible completions, the Page Zero experience can also show users the information without the user completing the entire query.

image

Building a Better Query

In this case, the user searches for “San Francisco Airport” and the most frequent intents associated to this query are “Maps”, “Weather”, “Parking” and “Arrivals”. For instance a click on “Weather” will produce a search page focused on the weather for SFO. We focus on helping the user select the right intent, and not worrying about the keywords that go along with it. Page Zero takes the frustration of not knowing the exact keywords out of the equation by summarizing the most common tasks and building the right query for the user.

image

Deep Learning

Intents are targeted to multiple domains, and these domains will keep expanding over time. In the last example, we searched for a university and we received very specific sub-intents such as “Admissions”, and “Ranking” in addition to the broader “News” and “Image” intents. Engaging with these tiles on Page Zero will lead to search results pages focused on the intent, which allow users to find the information they were looking for much quicker.

 

mit admin

 

 

In all of these examples, the Page Zero experience on Bing returns results which go beyond traditional search results by employing Satori in combination with multiple layers of machine learning that analyze the relationship between not only queries and entities, but also entities and documents. Using the power of Bing’s web index in combination with the Satori graph, we are able to select the most likely intents from thousands of intents expressed in billions of documents about billions of entities. And all of this happens faster than you blink.

People in the US can give this a try today at www.bing.com/preview.

Let us know what you think,

- Dan Marantz, Senior Program Manager, Bing