using interaction signals for job recommendation

Post on 23-Feb-2017

464 Views

Category:

Science

2 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Using Interaction Signals for Job Recommendation

Benjamin Kille, Fabian Abel, Balázs Hidasi, Sahin Albayrak| SIREMTI | 13 November 2015

Agenda

– Looking for a Job: now and then– Data Description– User Inquiry– Findings– Conclusion and Outlook

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 2/26

Problem Description

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 3/26

http

://pe

acec

hild

.org

/wp-

cont

ent/u

ploa

ds/2

015/

09/Y

outh

-une

mpl

oym

ent.j

pg

Traditional Method to Look for a Job

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 4/26

http

://w

ww

.bus

ines

srev

iew

aust

ralia

.com

/pub

lic/u

ploa

ds/la

rge/

larg

e_ar

ticle

_im

640_

new

spap

er_a

d_2.

jpg

Tends in Job Seeking

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 5/26

Use of print media decreasesProfessionals predominantly use

– Online job offer collectors– Online business networks

Trend leads to– Higher volume of job offers to

process for professionals– Higher volume of candidates to

deal with for employers– Reciprocal selection problem

Weitzel et al. (2015). Bewerbungspraxis 2015 – Eine empirische Studie mit 7000 Stellensuchenden und Karriereinteressierten im Internet.

http

://3.

bp.b

logs

pot.c

om/-T

m8S

-4m

JcdQ

/Tx2

Y-N

D1f

lI/A

AA

AA

AA

AC

cM/l6

fPkk

R9R

MI/s

1600

/info

rmat

ion_

over

load

_hyd

rant

.jpg

Reciprocal Selection

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 6/26

Professional:– reduce job offers to manageable size– remove irrelevant job offers– keep relevant job offers

Recruiter:– reduce list of candidates to

manageable pool– keep candidates with required skills– keep candidates likely to respond– remove candidates lacking necessary

skills

Ideally: match needs of both parties

How do we select job offers/candidates?

1. learn a modell representing professionals’ requirements

– curriculum vitae/skills

– location

– preferences

2. apply modell to available job offers

3. present suggestions to professionals

4. observe how professionals react

5. adjust modell to improve suggestions (repeat)

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 7/26

Types of FeedbackWe track users ...

... clicking on

... bookmarking

... replying to

suggested job offers

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 8/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 9/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 10/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 11/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 12/26

Which Type of Feedback should we use?

What can a click tell us?

What can a bookmark tell us?

What can replies tell us?

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 13/26

A User Inquiry

– How satisfied are users with their job recommendations?

– Collect ratings for job recommendations

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 14/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 15/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 16/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 17/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 18/26

What Type of Feedback Tells us Most?

Idea: Check which kind of Feedback correlates best with ratings:

– ratings ~ clicks

– ratings ~ bookmarks

– ratings ~ replies

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 19/26

Relation: ratings ~ clicks

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 20/26

clicks

ratings

1 2 3 4 5 6 7 8 9 10

12

34

5

Relation: ratings ~ bookmarks

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 21/26

Ratings for Bookmarked Jobs (μ = 3.6)

rating

Den

sity

0 1 2 3 4 5

0.0

0.1

0.2

0.3

0.4

0.5

Relation: ratings ~ replies

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 22/26

replies

ratings

1 2 3 4 5 6 7 8

12

34

5

Signal Comparison

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 23/26

clicks

ratings

1 2 3 4 5 6 7 8 9 10

12

34

5

Ratings for Bookmarked Jobs (μ = 3.6)

rating

Den

sity

0 1 2 3 4 5

0.0

0.1

0.2

0.3

0.4

0.5

replies

ratings

1 2 3 4 5 6 7 8

12

34

5

Conclusion and Outlook

feedback is necessary to improve recommendations

analysis of three signals:

– clicks à few clicks might be misleading

– bookmarks à filter bad suggestions; concentrate on medium preferences

– replies à most accurately reflect preferences

next steps

– implement a recommendation strategy that learns with replies

– A/B testing to verify suitability

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 24/26

Questions?

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 25/26

Using Interaction Signals for Job Recommendation | Benjamin Kille | SIREMTI 2015slide 26/26

Benjamin Kille (TU Berlin)

Competence Center Information Retrieval & Machine LearningInstitute of Commercial Information Technology and Quantitative Methods

benjamin.kille@tu-berlin.de@bennykille

http://crowdrec.eu

http://xing.com

top related