AI and the Future of Recruitment - Recruiting Community · Emb Deep NN Emb Deep NN CV Applications jobs Distance calculation Deep Learning Matcher Learn representation in shared space

AI and the

Future of

RecruitmentJakub Zavrel

About me

Jakub Zavrel

Founder of Textkernel (2001), since 2015 part of

CareerBuilder group. New venture: Zeta Alpha (2019)

R&D background in AI, Machine Learning & NLP since 1990.

Born 1970 in Prague, raised in Rotterdam, lives in

Amsterdam with wife and 2 teenage daughters.

Eclectic music tastes.

AI for People and Jobs / Semantic Recruitment / Labor

Market Analysis / Natural Language Processing / Machine

Learning / Search & Match / Enterprise Software

@jakubzavrel

[email protected]

I like programming, but I’m interested do take on more project

management responsibility

Is there a job in our organisation that better fits my degree?

I’d like to work on our mobile strategy. I’ve helped a friend

develop a mobile app.

I’d like to do more with my organisational

talent.

We are looking to hire:

An experienced tech team team lead

Labour Market:

a Language Gap

The ideal candidate has:

- min. 5yr of experience

- Certfied scrummaster

- Exp. w/iOS, Android

Completed academic studies

Computer Science or related

30% travel for customer presentations

In the next 5-10 years, AI will reach a level of near or super

human performance in many domains, including understanding

language and connecting supply and demand.

AI will fundamentally change the way people and jobs connect

in the global marketplace, and will remove market barriers

caused by language meaning variation and lack of

transparency.

What will this mean to recruiters and job seekers?

Why?

What do recruiters do?

Attracting

candidatesProcessing

candidatesSelecting

candidates

Chasing Line

Managers

Sourcing

Re-engagement

Employer Branding

Advertising

Referral Basic information

Scheduling

Data Entry

Screening

Scoring

Job offers

Contracts

Interviews

Assessment

WTF?!

Understanding

the job

AI Revolution: lowers the cost of repetitive cognitive tasks...

AI: The miracle algorithm?

Right candidate, right time, right price!

So what if you have technology that actually understands:

• the real requirements of the job

• the key factors to be successful

• the type of person most likely to succeed

• what your offer should be

So what do we mean by AI?

The power of machine learning

“Human-level control through deep reinforcement learning” V Mnih, et al. Nature, 2015

https://youtu.be/TmPfTpjtdgg

Understanding concepts and relationships

“Deep Visual-Semantic Alignments for Generating Image Descriptions”

Andrej Karpathy, Li Fei-Fei, et al. CVPR, 2015

thispersondoesnotexist.com- Generating Deep Fakes

“A Style-Based Generator Architecture for Generative Adversarial Networks”

Karras, et al. arXiv, 2018

AI learning to use tools?

“Improvisation through Physical Understanding: Using Novel Objects

as Tools with Visual Foresight”, Xie, et al. ArXiv, 2019

https://bair.berkeley.edu/blog/2019/04/11/tools/

https://techcrunch.com/2019/02/17/openai-text-generator-dangerous/

See: https://openai.com/blog/better-language-models/

February 14th, 2019.

Try it yourself: https://talktotransformer.com/

Ability to solve problems with unknown rules

What does Machine Learning give us?

Pattern discovery in volumes of data no human being is able to digest

What does Machine Learning give us?

Optimize solutions to problems with any value that we can define

What does Machine Learning give us?

What does Machine Learning give us?

Any human behavior for which we can collect sufficient data can be automated

by AI / Machine Learning

AI is Mainstream

Why now? Computing power, data, algorithms

Deep Learning

Deep Learning is the ability

for AI algorithms to automatically

learn meaningful patterns in

data using layered brain-like

structures called

Deep Neural Networks

Deep Learning: Hierarchically abstract information

• Use lots of unlabeled data• Deep multi-layered networks• Automatically finds relevant features

Understanding, connecting and analyzingpeople and jobs

Given a job, how do we find the relevant

candidates for the job among thousands or millions of CV’s?

Match!

Search!

CV Extraction

job=Java Developercity=Amsterdamlangskill=Germanexperience=7

CV MatchNormalizer

job=23branch=ITlangskill=DEexperience=7loc=Amsterdam

Vacancy MatchNormalizer

job=23branch=IT

langskill=DEexperience=5..10

loc=Amsterdam+10

Vacancy Extraction

job=Java Ontwikkelarcity=Amsterdamlangskill=Duitsexperience=7

Match!

Match! models construct the Search! Data Model out of extracted information.

The XML fits the Search! Data Model and is semantically enriched when INDEXING.

The result is a QUERY that fits the Search! Data Model. It is semantically enriched when executed

Data Model

job branch langskill location

experience

item 1

item 2

item 3

Previous Machine Learning Approach:

- Hidden Markov Models

- Conditional Random Fields

developermanager

engineer

CEO

CF

OCT

O

CO

O

2001 CEO at TextkernelInput:

Date Job - Company

Deep Learning for CVs and JobsWord embeddings

Recurrent Neural Networks (LSTM)

Extract! 4.O

Disrupting the playing fieldA qualitative leap through Deep Learning

Deep Learning

Textkernel

AC

CU

RA

CY

CV parsing

(competition)

job=Java

Developer

city=Utrecht

langskill=German

experience=7

job=23

branch=IT

langskill=DE

experience=7

loc=Utrecht

job=23

branch=IT

langskill=DE

experience=5..10

loc=Amsterdam+30

job=Java

Ontwikkelar

city=Amsterdam

langskill=Duits

experience=7

Match!

Search!

Data Model

job branch

langskill

location

experience

CV parsingSemantic

understanding

Semantic

understanding

Vacancy parsing

Intentions/Needs Intentions/Needs

Behavior:

Clicks, applies

Behavior:

Clicks, shortlist

Learning To Rank, Mine knowledge

Example of Matching

Turning the job description into a rich query

Learning from Feedback:

• Which candidate is the best?• Which criterion is more important?

• How about combination of criteria?

• How about domain biases?

• What if we can learn this automatically from feedback?

• Learning to Rank• Reorder a pool of candidates through machine learning

• E.g. Based on recruiter preference or past performance

Next Chapter:Deep Learning Matching

CV to Job Match: document embeddings

CV

Match!

query

Search!

JOBs

Job-title, it

skills,

experience

years,

keywords...

CV

JOBs

DLMatch!

Emb Deep NN

Emb Deep NN

CV

Applications

jobs

Distance

calculation

Deep Learning Matcher

Learn representation in shared space of CV and Vacancy texts such that Relevant CVs are close to the Vacancy and Irrelevant CVs are far.

SECURITY

OFFICER

SECURITY

OFFICER JOB

SECURITY

OFFICER JOB

REGISTERED

NURSE JOB

SECURITY

OFFICER

SECURITY

OFFICER JOB

SECURITY

OFFICER JOB

REGISTERED

NURSE JOB

Push far

Push close

SECURITY

OFFICER

Summary Efficient and organized surveillance professional with 15 years in security and safety compliance. Accomplished management

professional specializing in creating, launching and operating retail locations. Experience 10/2014-Present Security Officer US Security 04/2014-

01/2015 Security Officer Bristol Protective Service 08/2012- 09/2013 Metro Task Force Patrolled the facility and served as a general security

presence and visible deterrent to crime and rule infractions. Reported all incidents, accidents and medical emergencies to law enforcement.

Patrolled industrial and commercial premises to prevent and detect signs of intrusion and ensure security of doors, windows and gates.

Continuously monitored security cameras and fire, building and alarm systems. 08/2008-05/2009 Kitchen Director, Garden Day Care Center *

Collaborated extensively with interdisciplinary care team to meet the nutritional needs of each Senior. Established healthful and therapeutic meal

plans and menus. Encouraged clients and caregivers to follow recommended food guidelines for well-balanced diets. 08/2006-04/2008 Cashier,

Kmart * Maintained up-to-date knowledge of store policies regarding payments, returns and exchanges. Excelled in exceeding daily credit card

application. Created new processes and systems for increasing customer service satisfaction. Replenished floor stock and processed shipments

to ensure product availability for customers. 10/2005-01/2006 Sales Associate, Express Clothing Store * Computed sales prices, total purchases

and processed payments. Operated a cash register to process cash, checks and credit card transactions. Recommended merchandise based on

customer needs. Explained information about the quality, value and style of products to Influence customer buying decision. Education 01/2016-

05/2016 Essex County College High School Diploma

Query

ResultResult #1

Security Officer (New Jersey)

Join one of the fastest growing security companies in the U.S.! Since 1998, Sunstates Security has established a reputation for providing

excellent customer service and quality work environments while being recently recognized as one of the top 25 largest security providers in

America.

Result #2

Security Officer - Regular - Paterson, NJ

Observes and reports activities and incidents at an assigned client site, providing for the security and safety of client property and

personnel.Makes periodic patrols to check for irregularities and to inspect protection devices and fire control equipment.Preserves order and may

act to enforce regulations and directives for the site pertaining to personnel, visitors, and premises.Controls access to client site or facility through

the admittance process

Result #3

Security Officer - Regular - Summit, NJ

Observes and reports activities and incidents at an assigned client site, providing for the security and safety of client property and

personnel.Makes periodic patrols to check for irregularities and to inspect protection devices and fire control equipment.Preserves order and may

act to enforce regulations and directives for the site pertaining to personnel, visitors, and premises.Controls access to client site or facility through

the admittance process

DLM is

33%less likely to return irrelevant results

Intentions/Needs Intentions/Needs

I’d like to do more with my

organisational talent.

I’d like to work on our mobile strategy. I’ve helped

a friend develop a mobile app.

Is there a job in our organisation that better

fits my degree?

I like programming, but I’m interested do take on

more project management responsibility

Motivated and driven sales rep

We need an engineer with great customer

interaction skills

Somebody to spearhead our corporate

strategy

I want somebody to replace my best engineer that is

leaving at the end of the month

Match!

Behavior:

Clicks, applies

Behavior:

Clicks, shortlist

Assessments:Questionnaires,

personality, culture

Assessments:Questionnaires,

personality, culture

Jobseeker

AI AssistantRecruiter

AI Assistant

Intentions/Needs Intentions/Needs

I’d like to do more with my

organisational talent.

I’d like to work on our mobile strategy. I’ve helped

a friend develop a mobile app.

Is there a job in our organisation that better

fits my degree?

I like programming, but I’m interested do take on

more project management responsibility

Motivated and driven sales rep

We need an engineer with great customer

interaction skills

Somebody to spearhead our corporate

strategy

I want somebody to replace my best engineer that is

leaving at the end of the month

Match!

Jobseeker

AI AssistantRecruiter

AI Assistant

Requirements:

Services

Powers desktop + mobile apps

Chat bot capability

Messaging center

Market platform

PLATFORM

Functionality

Semantic Search & Match

Knowledge graph

S&D data (fusion with other sources)

User profiles

Language understanding

Conversational

Agents

• 24/7 service

• Massive outreach to

candidates

• Avoid black hole, keep

interaction going

• Pre-screen for interest

and availability, as well

as skills.

Conversational Search

• Conversational interfaces• Guiding, “do you mean”, etc

• Natural language search• Query understanding

• Chatbot interaction

AI: The miracle algorithm?

Right candidate, right time, right price!

Remember: it’s all measurable!!!

Fairness? Bias?

Explainability?

And remember this:

• Machines will not call in sick or complain about the work they do, but they will also not likely be able to enjoy it!

• So please do enjoy your job and celebrate your success! (while it lasts ;-)