a multidimensional expertise recommender tool
TRANSCRIPT
A multidimensional expertise recommender tool
German Sanchez-Hernandez, Jennifer Nguyen, Nuria Agell, Cecilio Angulo
June 24th, 2015
IntroductionState of the art
ArquitectureUser interface
Conclusions
Outline
1 Introduction
2 State of the art
3 Arquitecture
4 User interface
5 Conclusions
2 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Motivation and framework
Outline
1 Introduction
2 State of the art
3 Arquitecture
4 User interface
5 Conclusions
3 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Motivation and framework
Motivation and framework
www.projectcollage.eu
Creative learning is social, collaborative and peer based.
Expand and foster interaction among users with differentbackgrounds, opinions and levels of expertise→ improvement of creativity.
Find people with expertise in some areas(“right level of expertise” vs. “right person”)
4 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Outline
1 Introduction
2 State of the art
3 Arquitecture
4 User interface
5 Conclusions
5 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Related work of ERS
MITRE’s Expert Finder [Mattox et al., 1999]: experts intopics. Associations author-term in internal documents.Search by keywords.
NASA POPS [Grove and Schain, 2008]: filtering experts byorganisation, project or competency. Use of RDF, multipledatabases, semantic web.
IBM SmaillBlue [Lin et al., 2008; 2009]: expertise in emails andchat, mapping search strings to keywords.
INDURE FacFinder [Fang et al., 2008]: information fromfaculty profiles and homepages, all internal documentsindexed. Proximity of terms. Considers order, source and rankof source.
StrangersRS [Guy et al., 2011]: recommendation of people withsimilar interests but unfamiliar.
6 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Lacks
Just expertise
Information treatment:
Filtering candidatesParameters required (weights, thresholds)
→ Premature discards
7 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Lacks
Just expertise
Information treatment:
Filtering candidatesParameters required (weights, thresholds)
→ Premature discards
7 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Lacks
Just expertise
Information treatment:
Filtering candidatesParameters required (weights, thresholds)
→ Premature discards
7 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Lacks
Just expertise
Information treatment:
Filtering candidatesParameters required (weights, thresholds)
→ Premature discards
7 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Related work
Lacks
Just expertise
Information treatment:
Filtering candidatesParameters required (weights, thresholds)
→ Premature discards
7 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
Outline
1 Introduction
2 State of the art
3 Arquitecture
4 User interface
5 Conclusions
8 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
Arquitecture
9 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
ArquitectureProfiling
Profiling module
Candidates’ profiles
Access to Collage UserProfile Service
Offline massive updateOnline selective update
10 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
Profiles
Skill Profiling CER
Expertise Areas of knowledgeFour qualitative levels(none, high, medium, low)
Qualities/Subskills Other knowledge Specific tools
Proximity Ease to contactPhysical distanceHigh or low
Availability Current availabilityManual updateFour levels
11 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
ArquitectureInteraction
Interaction module
Interaction with the user
Translates preferences torequirements
ExplicitImplicit (user profile)
Controls for selecting finalexperts
12 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
ArquitectureIdentification
Identification module
Initial list of candidates
Access to profilesFeasible candidates
13 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
ArquitectureSelection
Selection module
Assessing each candidate
Aggregating assessments
Ranking and selection ofcandidates
14 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
Selection of candidatesAssessment
One assessment per requirement. Fuzzy distance between therequirement and the fulfillment.
Expertise: Ae(ps , l) = min(ps ,l)l .
Subskill: Aq(ps) = ps .
Proximity: distance between user and candidate.
Ap(pd) =
{1− pd−md
Md−mdif high proximity is required,
pd−mdMd−md
otherwise.
Availability: distance to (high) required availability
Aa(pa) = pa. (1)
(ps , pd and pa are related to the profile of the candidate; l is the requiredlevel of expertise; md and Md are the minimum and maximum distancesbetween departments.)
15 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
ArquitectureCER
Selection of candidatesAggregation and Ranking
Operator: OWA.Weights: linguistic quantifier
wi = Q
(i
n
)− Q
(i − 1
n
), i = 1, . . . , n.
“most of”: Q(r) = r12 .
16 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Outline
1 Introduction
2 State of the art
3 Arquitecture
4 User interface
5 Conclusions
17 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Current UIInput form
18 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
New UIInput form
19 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Current UIResults
20 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
New UIResults
21 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Outline
1 Introduction
2 State of the art
3 Arquitecture
4 User interface
5 Conclusions
22 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Conclusions and Next Steps
Expertise recommender to find the right expert.
Right level of expertise vs. right person.
Additional information used: qualities, proximity, availability.
Integration into an existing platform (end users’ affinity space,Moodle).
Information available in end users’ systems.
Improving prototype.
Team forming.
23 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender
IntroductionState of the art
ArquitectureUser interface
Conclusions
Conclusions and Next Steps
Expertise recommender to find the right expert.
Right level of expertise vs. right person.
Additional information used: qualities, proximity, availability.
Integration into an existing platform (end users’ affinity space,Moodle).
Information available in end users’ systems.
Improving prototype.
Team forming.
23 / 24 German Sanchez-Hernandez JARCA 2015: Expertise Recommender