Social Search in a Professional Context

This 2013 ACM Conference on Information and Knowledge Management (CIKM) presentation discusses data-driven user behavior modeling for professional search on LinkedIn. It emphasizes the differences between professional and web search, focusing on entity-oriented search, query tagging, and the importance of user networks in search relevance. It highlights methods for personalizing search results and improving query understanding to better connect talent with opportunities.

Transcript

1. Recruiting Solutions Recruiting Solutions Recruiting Solutions Social Search in a Professional Context Daniel Tunkelang LinkedIn, Head of Query Understanding 1 Daniel Workshop on Data-driven User Behavioral Modeling and Mining from Social Media
2. LinkedIn connects talent to opportunity. 2 Search enables the participants in the economic graph to find and be found.
3. Overview Why do people search in a professional context? How do we help people search in a professional context? Next play? 3
5. Scenario 1: Pleased to meet you! 5
6. People search isn’t the same as web search. 6
7. LinkedIn works hard to make it effortless. 7
8. Even harder to reduce user effort to a few chars. 8
9. Searchers use what they know to find people. 9
10. Not all navigational queries are name searches. 10
11. Scenario 2: Looking for new opportunities. 11
12. Lots of jobs in DC for data scientists. 12
13. My connections can help me get a $100k+ job. 13
14. Apply, contact the recruiter, or seek a referral. 14
15. Scenario 3: I know what I want when I see it. 15
16. Another year, another CIKM industry event. 16
17. We’ll need student volunteers, too. 17
18. And some sponsors! 18
20. LinkedIn’s focus: entity-oriented search. 20 Company Employees Jobs Name Search
21. Query tagging: key to query understanding. §  Using human judgments to evaluate tag precision. –  Extremely accurate (> 99%) for identifying person names. –  Harder to distinguish company vs. title vs. skill (e.g., oracle dba). §  Comparing CTR for tag matches vs. non-matches. –  Difference can be large enough to suggest filtering vs. ranking: 21
22. Query Tagging: An Example 22
23. Detecting navigational vs. exploratory queries. Pre-retrieval §  Sequence of query tags. Post-retrieval §  Distribution of scores / features. 23 Click behavior §  Title searches >50x more likely to get 2+ clicks than name searches.
24. Navigation vs. Exploration: Behavior Patterns §  Exploratory searches leads to ~5x more clicks per search than navigational searches. §  Clicks on 2nd-degree connection more than 2x as likely to lead to invitation from exploratory vs. navigational search. §  For navigational queries, 1st degree > 2nd degree > … §  For exploratory queries, 2nd and 3rd degree > 1st degree. 24
25. Query expansion for exploratory queries. 25 software patent lawyer Query expansions derived from reformulations. e.g., lawyer -> attorney
26. LinkedIn search is personalized. 26 kevin scott
27. But global factors matter. 27
28. Relevant results can be in or out of network. 28 §  Searcher’s network matters for relevance. –  Within network results have higher CTR. §  But the network is not enough. –  About two thirds of search clicks come from out of network results.
29. Personalized machine-learned ranking. 29 §  Data point is a triple (searcher, query, document). –  Searcher features are important! §  Labels: Is this document relevant to the query and the user? –  Depends on the user’s network, location, etc. –  Too much to ask random person to judge. §  Training data has to be collected from search logs.
30. How to train your model. 30 §  Train simple models to resemble complex ones. –  Build Additive Groves model [Sorokina et al, ECML ’07], which is good at detecting interactions. §  Build tree with logistic regression leaves. §  By restricting tree to user and query features, only regression model evaluated for each document. β0 +β1 T(x 1 )+...+βn x n α0 +α1 P(x 1 )+...+αn Q(x n ) X2 =? X10 < 0.1234 ? γ0 +γ1 R(x 1 )+...+γn Q(x n )
33. Make search truly entity-centric. 33 results results
34. Use the search box to surface task intent. 34 I am… looking for a job… at LinkedIn in Fiji trying to hire… software engineers web developers interested in learning about… Hadoop NoSQL
35. It takes two to connect talent to opportunity. 35
36. LinkedIn: connecting talent to opportunity. 36 Search: enabling the participants in the economic graph to find and be found.
37. Thank you! 37 238,
38. Want to learn more? §  Check out http://data.linkedin.com/search. §  Contact me: dtunkelang@linkedin.com http://linkedin.com/in/dtunkelang §  Did I mention that we’re hiring? J 38