Learning to Rank Personalized Search Results in Professional Networks
Download as PPTX, PDF2 likes1,191 views
1) The document discusses personalized search solutions for professional networks like LinkedIn, including augmenting short queries with user profile data, calculating skill reputations to find relevant jobs, and using a personalized federated search model that considers user intent and signals from different content verticals. 2) It describes challenges like skill sparsity and outliers, and approaches used to estimate skill reputation scores and infer missing skills based on collaboration. 3) The conclusions are that text matching is not enough, and personalized learning-to-rank which considers semi-structured user data, behavior, and collaborative filtering is crucial for search.
![Personalized Job Search
▪ Short and vague queries
–“San francisco”, “microsoft”
–Augment queries with searcher information
▪ Skill Homophily [Li, Ha-Thuc et al. KDD’16]
–“Classic” homophily: People tend to connect with similar ones
–Skill homophily: People tend to apply for jobs requiring similar skills
–Skills in job descriptions
5](https://image.slidesharecdn.com/sirip2016-160721103111/85/Learning-to-Rank-Personalized-Search-Results-in-Professional-Networks-5-320.jpg){kind=icon}
![Skill Reputation Scores [Ha-Thuc et al. BigData’15]
8
▪ Decision Maker: searcher
▪ Record: Professional
career
▪ Skill reputation: member
expertise on a skill
▪ Judgment: Hire?](https://image.slidesharecdn.com/sirip2016-160721103111/85/Learning-to-Rank-Personalized-Search-Results-in-Professional-Networks-8-320.jpg){kind=icon}
![Personalized Federated Model [Arya, Ha-Thuc et al. CIKM’15]
▪ Searcher intent
– Mine searcher profiles and past behavior to infer intent
▪ Title recruiter -> recruiting intent
▪ Search for jobs -> job seeking intent
– Machine-learned models predict member intents:
▪ Job seeking
▪ Recruiting
▪ Content consuming
12](https://image.slidesharecdn.com/sirip2016-160721103111/85/Learning-to-Rank-Personalized-Search-Results-in-Professional-Networks-12-320.jpg){kind=icon}
Learning to Rank Personalized Search Results in Professional Networks
- 1. Recruiting SolutionsRecruiting SolutionsRecruiting Solutions Learning to Rank Personalized Search Results in Professional Networks Viet Ha-Thuc and Shakti Sinha SIGIR Industry - 2016 1
- 2. 2 ● Dual Roles of Search ○ Enable talent discover opportunity ○ Help companies to search for the right talent
- 3. Unique Nature of LinkedIn Search ▪ Heterogeneous sources – People, jobs, companies, slideshares, members’ posts, groups ▪Support many use-cases – Recruiting, connecting, job seeking, research, sales, etc. ▪Deep Personalization 3
- 4. Overview 4 Query Federated Search Spell Correction Query Tagging Intent Prediction People Companies Federated Search Name Title Skill Jobs
- 5. Personalized Job Search ▪ Short and vague queries –“San francisco”, “microsoft” –Augment queries with searcher information ▪ Skill Homophily [Li, Ha-Thuc et al. KDD’16] –“Classic” homophily: People tend to connect with similar ones –Skill homophily: People tend to apply for jobs requiring similar skills –Skills in job descriptions 5
- 6. Member Skills ▪ Skills – 35K+ standardized skills – Represent professional expertise ▪Challenges – Sparsity – Outlier skills ▪Approach: skill reputation 6
- 7. Reputation Information a decision maker uses to make a judgment on an entity with a record (*) 7 (*) “Building web reputation systems”, Glass and Farmer, 2010
- 8. Skill Reputation Scores [Ha-Thuc et al. BigData’15] 8 ▪ Decision Maker: searcher ▪ Record: Professional career ▪ Skill reputation: member expertise on a skill ▪ Judgment: Hire?
- 9. Estimating Skill Reputation 9 ● Remove outlier skills ● Infer missing ones
- 10. Overview 10 Query Federated Search Spell Correction Query Tagging Intent Prediction People Companies Federated Search Name Title Skill Jobs
- 11. ▪ Why do we need this? 11 Personalized Federated Search - Motivation
- 12. Personalized Federated Model [Arya, Ha-Thuc et al. CIKM’15] ▪ Searcher intent – Mine searcher profiles and past behavior to infer intent ▪ Title recruiter -> recruiting intent ▪ Search for jobs -> job seeking intent – Machine-learned models predict member intents: ▪ Job seeking ▪ Recruiting ▪ Content consuming 12
- 13. Calibrate Signals across Verticals ▪ Verticals associate with different intents 13 People Result Job Result Group Result Recruiting Intent Job Seeking Intent Content Consuming Intent
- 14. Calibrate Signals across Verticals ▪ Verticals associate with different intents 14 People Result Job Result Group Result Recruiting Intent Job Seeking Intent Content Consuming Intent
- 15. Calibrate Signals across Verticals ▪ Verticals associate with different intents 15 People Result Job Result Group Result Recruiting Intent Job Seeking Intent Content Consuming Intent
- 16. Take-Aways ▪ Text match is still important but not enough ▪ Go beyond text ▪Semi-structured data ▪Behavioral data ▪ Collaborative filtering works for skill reputation ▪ Personalized Learning-to-Rank is crucial 16
- 17. References ▪“Personalized Expertise Search at LinkedIn”, Ha-Thuc, Venkataraman, Rodriguez, Sinha, Sundaram and Guo, BigData, 2015 ▪“Personalized Federated Search at LinkedIn”, Arya, Ha- Thuc and Sinha, CIKM, 2015 ▪“How to Get Them a Dream Job?”, Li, Arya, Ha-Thuc, Sinha, KDD, 2016 17