Database Reference
In-Depth Information
Gutierrez:
Why is query understanding an important problem to tackle for
LinkedIn?
Tunkelang:
Search is what allows LinkedIn's hundreds of millions of mem-
bers to find each other and be found. LinkedIn members did over 5.7 billion
professionally-oriented searches on the platform in 2012, and that number
has kept growing since then. Search is one of the most important ways that
LinkedIn's members engage with our platform.
It's our job to make sure that our members find what or who they are looking
for. And because the search experience on LinkedIn is highly personalized,
we face unique challenges in delivering quality results to our members. Our
search engine needs to take into account who you are, who you know, and
what we know about your network to help you find what you're looking for.
Query understanding has been an important problem for some time. I started
thinking about it when I was at Endeca, working with faceted search and semi-
structured data sets. Since LinkedIn is a poster child for both, it was natural to see
how better query understanding could improve the search experience. I'd been
dabbling in this area for a while, and in 2013, I decided to focus on it exclusively.
Gutierrez:
Why is query understanding interesting to you?
Tunkelang:
Query understanding offers the opportunity to bridge the gap
between what the searcher means and what the machine understands. Instead
of tackling the squishy, subjective problem of how relevant a piece of content
is to the searcher, it focuses on the more objective problem of establishing an
unambiguous information need, so we can figure out which content is relevant
at all. Also, we're able to improve the language of communication between
the searcher and the machine, which is an exciting development in human-
computer interaction.
Gutierrez:
Why is search relevance an important problem to tackle for LinkedIn?
Tunkelang:
Search is a pillar of LinkedIn's platform—it's what enables our
300M+ members to find and be found. But of course the search results have to be
relevant. Our members perform billions of searches, and each of those searches is
highly personalized based on the searcher's identity and relationships with other
professional entities in LinkedIn's economic graph. It's a challenging problem in
several dimensions, and solving it delivers enormous values to our members.
Gutierrez:
What have you been you working on this year?
Tunkelang:
As its name suggests, the Query Understanding team has been
working on understanding queries—specifically, the queries our members
issue when they search on LinkedIn. We look at understanding queries before
deciding which results to retrieve and score. For example, if someone searches
for “daniel linkedin”, we can figure out that “daniel” is a person's name and
“linkedin” is a company, so we only retrieve results corresponding to people
