Recruiting · Production

LinkedIn extracts skills from unstructured content using AI to power the LinkedIn Skills Graph

The problem

Skills across LinkedIn's platform are embedded in unstructured content — member profiles, job postings, and learning courses — and not consistently listed in structured fields, making comprehensive skills-based matching difficult to build at scale.

Workflow diagram · grounded in source
1
Content sources trigger extraction
trigger
“we use AI to extract skills from various content sources across LinkedIn and map these skills to our Skills Graph”
2
Skill segmentation
ai_action
“Before extracting any skill, we first parse the raw input into a well-formed structure. A job posting, for example, may have sections for "company description," "responsibilities," "benefits," and "qualifications." Meanwhile, a resume wi…”
3
Skill tagging
ai_action
“a skill tagger identifies the mentions of skills in the text. The skill tagger can perform token-based skill matches and also perform semantic-based skill matches from short sentences/phrases”
4
Skill expansion via Skills Graph
ai_action
“The skill expansion relies on our Skills Graph to query for relevant skills in the same skill group or skills that share structural relationships, such as parent skills, children skills, and sibling skills”
5
Multitask cross-domain scoring
ai_action
“a multitask scoring runs to identify each content piece and skill candidate pair. The multitask scoring model contains two parts, a shared module and a domain-specific module”
6
Skills served to downstream systems
output
“those extractions can be used by LinkedIn members, products, and AI systems (including search, recommendations, feed ranking, Jobs You May Be Interested In, Job Search, Recruiter Search, and many others)”
7
Feedback loops improve models
feedback_loop
“we build feedback loops directly into online Job Posting and member profiles to help with AI model iterations”
Reported outcome

LinkedIn's multitask AI skill extraction system improved job recommendation, job search, and skills matching results across multiple A/B tests — with gains in predicted confirmed hires, qualified applications, and PPC revenue.
Knowledge Distillation enabled 80% model size reduction for nearline serving while meeting strict latency requirements.

Reported metrics
model size reduction via Knowledge Distillation80%
Global profile edits per second processedapproximately 200
per-message processing latency SLAunder 100 milliseconds
Job Recommendation: Member Job Applicants and Offsite Apply Clickers+0.1391%
Show all 12 reported metrics
model size reduction via Knowledge Distillation80%
global profile edits per second processedapproximately 200
per-message processing latency SLAunder 100 milliseconds
Job Recommendation: Member Job Applicants and Offsite Apply Clickers+0.1391%
Job Recommendation: Predicted Confirmed Hires+0.4606%
Job Search: Job Sessions+0.1468%
Job Search: PPC revenue+0.7577%
Job Search: Engagements+0.2271%
Job Member Skills Matching: Qualified Applications+0.87%
Job Member Skills Matching: Qualified Application Rate+0.40%
Job Member Skills Matching: Predicted Confirmed Hires+0.24%
Job Member Skills Matching: Applicants and Apply Click Counts+0.48%
Reported stack
Multilingual BERTLLMTransformerKnowledge DistillationSparkSamza-BEAM
Source
https://www.linkedin.com/blog/engineering/skills-graph/extracting-skills-from-content
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

LinkedIn's multitask AI skill extraction system improved job recommendation, job search, and skills matching results across multiple A/B tests — with gains in predicted confirmed hires, qualified applications, and PPC…

What tools did this team use?

Multilingual BERT, LLM, Transformer, Knowledge Distillation, Spark, Samza-BEAM.

What results were reported?

model size reduction via Knowledge Distillation: 80%; Global profile edits per second processed: approximately 200; per-message processing latency SLA: under 100 milliseconds; Job Recommendation: Member Job Applicants and Offsite Apply Clickers: +0.1391% (source-reported, not independently verified).

How is this recruiting AI workflow structured?

Content sources trigger extraction → Skill segmentation → Skill tagging → Skill expansion via Skills Graph → Multitask cross-domain scoring → Skills served to downstream systems → Feedback loops improve models.

Related recruiting cases