Daniel Tunkelang on Enterprise Search

Daniel TunkelangDaniel Tunkeland, Head of Query Understanding at LinkedIn, talked earlier this week at the Enterprise Search Summit about  “Enterprise Search: How Do We Get There From Here?

Here are the slides and below is Tunkelang’s summary of the presentation:

We dream of breaking down the communication barriers between people and machines. We want to talk with our computers like on Star Trek — better yet, to communicate using mind-reading devices. We long for Sir Tim Berners-Lee‘s vision of “a [semantic] web of data that can be processed by machines”.     

But today we are far from that fantasy. At Google, which epitomizes the success of web search, their own VP Udi Manber characterized their in-house search as “not that good“. For all the talk of moving beyond the “10 blue links” interface, we still follow that paradigm in most of our search experiences. And the idea of semantic, structured data is still more promise that reality.

So how do we get there from here?

I offered three baby steps on the path to utopia:

1. Exercise common sense.

There are lots of easy things we can do to make search better. Focus on head queries. If you have a long tail, then segment your queries and focus on the key segments (e.g., at LinkedIn those include name searches and job title searches). Take a lesson from online retail: focus the search experience on entities and categories — even in cases where you don’t have the content user is looking for. Finally, identify unsuccessful searches and use analytics to drive triage (I give an example of how we analyzed no-results name searches at LinkedIn).

2. Show some humility.

“You just ask them?” Those words come from the late Nobel laureate Richard Feynman in a somewhat different search context, but they apply perfectly to the world of enterprise search. Rather than try to read your users’ minds, focus on way to solicit their intent. Recognize ambiguity and ask for clarification. Clarify the user’s intent and then offer opportunities for the user to refine it — ideally using faceted search. Finally, recognize thatnot all queries are created equal in difficulty and adapt the search experience accordingly.

3. If all else fails, cheat.

On a South Park episode, Cartman tells a class of inner-city children that “it’s ok to cheat, as long as you cheat your way to the top.” I’m not sure I’d like my daughter to get her values from South Park, but cheating is sometimes the best way to improve the search experience. For example, you may be able to design an experience that doesn’t require search — as Apple has done with its online store. You can crowd-source curation by letting users create named shortcuts. You can obtain entities by recycling data, bootstrapping on small amount of labeled documents or even borrowing labels from search logs. Or you may be better of solving a different problem, such as moving more of the search experience to type-ahead.

Finally, I noted that technology alone won’t solve our problems. We have to make search experience a priority and invest time and effort to make our users happy.

I’m looking forward to the future of human-computer interfaces. But there’s a lot we can do with science and technology we already have. So let’s make some of that future happen today!

About GilPress

I launched the Big Data conversation; writing, research, marketing services; http://whatsthebigdata.com/ & http://infostory.com/
This entry was posted in Enterprise Search, LinkedIn, Search. Bookmark the permalink.

Leave a Reply