Exclusive: Interview with Sriram Sankar – LinkedIn Economic Graph

KDnuggets talks with Sriram Sankar, Principal Staff Engineer at LinkedIn about LinkedIn’s “Economic Graph”, Entity-Oriented Search, and the biggest challenges towards delivering relevant, personalized search results.

Sriram SankarSriram Sankar is a Principal Staff Engineer at LinkedIn, where he is leading the development of their next-generation search infrastructure. Before that, he led Facebook’s search quality and ranking efforts for Graph Search. He previously worked at Google on search quality and ads infrastructure and held senior technical roles at VMware, WebGain, and Sun. He was a key contributor to Unicorn, the index powering Facebook’s Graph Search, and developed JavaCC, the leading parser generator for Java. He is a graduate of the Indian Institute of Technology in Kanpur. LinkedIn

Here is my interview with him:

Anmol Rajpurohit: In the context of search, how do you differentiate LinkedIn's "economic graph" from Google's "knowledge graph" and Facebook's "relationship graph"?

Economic GraphSriram Sankar: LinkedIn and Facebook are both social networks – they both contain members with an identity, and contain many other entities useful to these members – companies, jobs, etc. in LinkedIn and pages, photos, etc. on Facebook. The members and other entities have relationships with each other resulting from actions by members – connections, likes, mentions, etc. Additional relationships can be inferred through data analysis. This is the basis of LinkedIn’s economic graph, or Facebook’s social graph.

Where they differ is that LinkedIn’s graph is more about professional relationships while Facebook’s is more about social relationships. Something unique about LinkedIn’s data is that it is mostly public which allows for a richer member experience and more valuable insights.

Google’s knowledge graph on the other hand is based on information gathered from a variety of sources and then enhanced through additional inferences.

AR: Can you explain the term "entity-oriented search" and its relevance?

SS: Search within data that is structured and connected in inherently entity oriented – for example, the nodes in the LinkedIn economic graph and the edges between them represent entities.

AR: You have been involved with search quality and ranking since a long time, including your work at Google and Facebook. What do you see as the biggest challenges towards delivering relevant, personalized search results?

LinkedIn Search SS: Search at LinkedIn is one piece of a fuller product offering – the LinkedIn experience. To do a good job with search at LinkedIn requires this realization and working with the rest of the teams at LinkedIn closely to deliver a better overall LinkedIn experience. On the other hand, the goal of search engines like Google, Bing, etc. are to get the searcher to a good results page outside their domain as quickly as possible.

What this means is that the approach to search relevance, search personalization, and finally the definition of search success (e.g., via metrics) needs to be very different.

In comparison with other social networks, search at LinkedIn is a much more fundamental piece of the LinkedIn experience. And making search better can go a long way in making the overall LinkedIn experience better.

AR: What have been the biggest infrastructure challenges towards serving the billions of searches from 277M+ LinkedIn members?

SS: While these numbers do feel large, the real challenges at LinkedIn come from the diversity of the different kinds of searches performed and the sophisticated nature of some of the searches. Recruiters hit our system hard with very complex queries and are looking for lots of results. Our infrastructure challenges come from having to deal with a variety of these complex queries.

AR: LinkedIn seems to have been hiring aggressively in Data Science arena. What key characteristics do you look for when evaluating applicants for opportunities on your team?
LinkedIn Hiring
SS: At the end of the day, we want our team to have a strong intuition for what our end users are looking for and a similarly strong intuition for what it takes to build such features into our search engine. And the willingness to take calculated risks. Obviously we expect a good data science background – familiarity with Hadoop, machine learning, etc.

AR: What was the last book that you read and liked? What do you like to do when you are not working?

SS: It has been a while since I read a serious book – there is so little time left after everything else. Furthermore, there is a lot of very interesting and diverse reading material online that compensate. Most of my free time is spent with my family and friends. And a little bit of exercise – that includes a weekly non-negotiable game of racquetball!