User Centered Search

Mobile Search: From Document Search to User Centered Search

Scout
6 min readJun 15, 2015

Bento Labs is a search company. People often ask us what that means and how search is different in the mobile world. Without going into all the details of how our product is implemented, we want to share how we think about search in a mobile world.

First some background on Web Search

In the late 90’s as the web grew exponentially there arose a need to be able to rapidly search through the different kinds of web pages on the web. We will call this “Document Search”. Document search involves:

  • Determining what documents need to be crawled
  • Crawling and downloading the documents
  • Indexing the documents by keywords
  • Building a system in which the user can type in the keywords and “relevant documents” can be surfaced to the user.

Relevance of documents to a keyword query is determined by matching the keywords against words in the document in multiple ways. The most famous ranking system, however, didn’t need query keywords at all; Google’s “Pagerank” assigned “relevance” to a particular document based on how many other “important” documents link to it. Armed with Pagerank, Google focused on building a generic search engine that didn’t need to understand the contents of the document it was indexing. Google therefore focused on having the most number of documents in its index and building a speedy search over them. Google’s obsession with speed is well documented and led to some incredible features like Google Instant

While Google was busy ingesting all the documents on the Web and building a generic search engine, a number of vertical search engines came about that chose to focus on exactly one particular content type and perfect it. Zillow, Yelp, SeatGeek, Spotify, HotelTonight, Amazon etc are great examples of this. As we rethink mobile search, we seek to utilize these vertical search engines to create a good experience for the user.

Mobile Search: Anticipate user needs at all times

Intelligence is the art of good guesswork

H. B. Barlow

Mobile apps are architected differently than websites and content inside them is not easily crawlable by a traditional search engine. This is one of the main reasons discovery of apps and app content on mobile is hard. Various deep linking schemes have emerged that try to link crawlable web content to mobile content but the effort feels piecemeal since it tries to replicate document search on mobile instead of taking advantage of the different computing paradigm available on mobile.

On mobile, while we may not know about documents inside apps, we know a lot more about users. The phone is a far more capable device than the desktop computer ever was, and most importantly it is with users 24/7. Phones have the ability to sense, infer, and maintain the user’s behavior history and combined with explicit user preferences can be used to create an interest graph of the user. This interest graph combined with the user’s context, can help us anticipate user need and PUSH relevant content from apps to the user such that it increases discovery of content from inside the apps.

With Bento we introduce User centered search to fundamentally rethink discovery in the era of mobile apps. User centered search relies on determining user intent and matching it with content from other apps. We take advantage of the fact that there exist domain specific search engines (or apps) that can do a really good job of finding information for the user. By constructing a framework for maintaining and sharing context specific queries with such apps we can have the phone’s home-screen be a contextually relevant stream of content from apps that the user may use.

If the above image doesn’t load click here or here

In this example, the user is at the gym and based on what Bento knows about the user, it estimates that at this moment in time, the user might be interested in music, news, and a cab to get back home. It then packages this information and shares it with relevant apps that may have a good “search result” for the user. Apps then respond with results for the user, which are then then ranked by Bento. Apps need not be installed on the user’s phone to be able to respond to the user’s query. Such a stream of configured cards will be available at all times on the phone’s home-screen, which would thus becomes the nerve center of the phone.

Benefits of User Centered Search enabled by Bento

Zero configuration apps + Instant personalization

Cards are functional apps that have been configured by Bento for the user. In the example above, the SoundCloud card creates a playlist of “Electronic” music because Bento told it that the user likes to listen to “Electronic” music. If the user were to download the SoundCloud app, it too could be instantly personalized via Bento.

Better discovery: Preview apps before downloading them

Apps can construct relevant cards for the user and the user gets to try the app before the user decides to download the app.

Do away with the distinction of a web app vs mobile app

Want to watch a video? Get a cab? A user shouldn’t care if there is an app or a website behind a service; such cards help drive the distinction away.

Information at fingertips at all times

Bento pushes relevant information to the user at all times. As the user uses Bento, Bento gets better at predicting what the user needs and the user has to do less because the phone does more.

Data visibility

Since Bento maintains updated information about the user and shares that information with apps, users will have one place to get full visibility into that data and be able to modify it as needed.

Privacy: Intelligent defaults + Opt in + User accounts

The most obvious question in the scheme above is what information about the user is being shared with these other services and how does the user control what is being shared.

In a world of zero configuration apps which needn’t necessarily be downloaded, a simple framework for thinking about privacy could be:

Intelligent defaults

Bento shares certain non-PII (Personally Identifiable information) information about the user with relevant apps. For example imagine that you are taking a trip to London and you don’t know what a good cab app for London is. As soon as you land in London, Bento can construct a Card(search result) for “Downtown London” with an estimated cab fare from a cab app like Hailo. The Hailo app still doesn’t need to know where exactly you are going, or even who you are, but sharing some information about your itinerary may be a way for Hailo to suggest a price and get you interested in trying the app.

Opt in

If you then decide to give Hailo more information about your destination you can go ahead and do that. This information may or may not be PII

User Account

At some point you have to create an account inside Hailo to share PII information. Bento will simplify that process and maintain the password in some sort of password manager so you can try many such services with very little effort.

Identity in this world

When the user has chosen to authenticate with a particular service, Bento will be able to share a lot more context than when not authenticated with the service — at the same time it can also pull in more personalized content for the user that the app may have available. A good app account manager holds the key to user managing passwords across services. Much like the way the user puts in all their credit cards in a mobile wallet and then uses the mobile wallet to pay for services, a good app account manager can make authentication simple in this world.

Conclusion

Phones are superior devices to desktop computers and there is absolutely no reason for us to accept a worse discovery experience on phones. Because of the reach that phones have been able to achieve, this problem is truly important. By maintaining the interest graph of the user and using context, Bento matches the user to services and apps that the user may use at any given time. Bento provides a framework which blurs the distinction between a web app and a mobile app and allows the user to experience apps before deciding to download them. In this age of context, information discovery should feel organic and seamless. Bento is building towards that future.

Bento is currently available only for Android and is in beta. You can join the beta here.

--

--