Morteza Zihayat
Mehdi Kargar
Aijun An
York University, Toronto, Canada
Two-Phase Pareto Set Discovery for
Team Formation in Social Network
Overview
• Team Formation in Social Networks
• Communication Cost
• Personnel Cost
• Expertise level
• Approximation Algorithm
• Empirical Results
• Conclusion
2
WIC’14 Two-Phase Pareto Set Discovery for ...
Team of Experts
• Given a social network,
find a team of experts that
minimizes the
communication cost, the
personnel cost and the
Expertise cost in order to
complete a project.
• Clearly, our problem is a
three criteria optimization
problem.
3
WIC’14 Two-Phase Pareto Set Discovery for ...
Team of Experts
• Project: set of required skills
• Expert: an individual with a specific skill-set
• Social Network: represents strength of
relationships (the degree of collaboration
between any two experts).
• For example: LinkedIn, DBLP and …
4
WIC’14 Two-Phase Pareto Set Discovery for ...
Team of Experts
• Feasible Team of Experts: Given a set of experts and a
project that requires a set of skills {s1, s2, . . . , sp}, a team of
experts is a set of p skill-expert pairs:
{(s1, cs1), (s2, cs2) , . . . , (sp, csp) },
where csk is an expert having skill sk for k = 1, . . . , p.
• A skill-expert pair (sk, csk)
• expert csk is responsible for skill sk in the project.
5
WIC’14 Two-Phase Pareto Set Discovery for ...
 How to make sure that the experts can communicate together?
 What about the personnel cost?
 How to make sure that the experts are the best experts for the
required skills (Expertise cost).
Communication Cost
• An experts social network is modeled as an undirected and
weighted graph G.
• For the following team of experts regarding graph G
• The sum of distances of a team of experts
• The sum of the shortest distances between the experts responsible for
each pair of skills.
6
WIC’14 Two-Phase Pareto Set Discovery for ...
Personnel Cost
• For the following team of experts regarding graph G
• An expert c is paid k × t(c)
• k is the number of skills the expert is responsible for in the project.
• More skills the expert uses → More responsibility or
tasks → More money he/she is received
7
WIC’14 Two-Phase Pareto Set Discovery for ...
Expertise Cost
• For the following team of experts regarding graph G
• Where is the expertise cost with respect to the skill
and is reversely proportional to the expertise level of the
expert.
8
WIC’14 Two-Phase Pareto Set Discovery for ...
Three Objective Team Formation
Problem
• In three objective optimization problem:
• There does not exist an answer that optimizes all the
criteria simultaneously.
• Instead of finding one single answer, we can find a set
of teams that are not dominated by others.
• Dominance Relation: Given two feasible teams T and T’,
T dominates T’ if and only if CC(T)≤CC(T’), EC(T) ≤ EC(T’)
and PC(T) ≤ PC(T’). It is denoted as
• Efficient Team: Team T is efficient (also called Pareto-
optimal) for a project P if and only if there does not exist a
feasible team T’ for P such that
9
Efficient Teams Categorization
• The efficient teams can be categorized into two main
types:
• Supported Efficient Teams: can be obtained by solving the
following weighted sum single-objective problem
• We denote this weighted-sum single objective problem as
WeightedSingleObjλ
• Non-supported efficient team. An efficient team that is not an
optimal team of any WeightedSingleObjλ problem.
10
Pareto Front Team Formation Problem
• Given a project P and a graph G representing the network of
experts, the Pareto front team formation problem is finding the
set of all efficient teams from G for P.
• We present a two-phase method for finding the set of efficient
teams in graph for a project.
• Lexicographic Optimal Team: Feasible team T is a lexicographic
optimal team with respect to a lexicographic order of objectives
(cost1,cost2,cost3) if for any other feasible team T’:
• Cost1(T) < Cost1(T’)
• Or, Cost1(T) = Cost1(T’) but Cost2(T) < Cost2(T’)
• Or, Cost1(T) = Cost1(T’) and Cost2(T) = Cost2(T’) and
Cost3(T) < Cost3(T’)
11
Overview of algorithm
12
Finding LOT while Optimizing Communication
Cost
• Finding the optimal team while minimizing the
communication cost is proved to be NP-hard
• We find a LOT with respect to the lexicographic order of
(CC,EC,PC)
• Since minimizing the communication cost is NP-hard, this
algorithm finds the best team with 2-approximation
13
Finding LOT while Optimizing Expertise Level
• We propose a method for finding a LOT with the minimum
expertise cost
• This algorithm finds an exact LOT in polynomial time
• Its time complexity is O(m2
× n) where m is the number of
experts in G and n is number of required skills for project
P.
• The similar algorithm is applied for finding LOT while
optimizing Personnel cost. The only difference is using
personnel costs instead of expertise costs.
14
Finding Supported Efficient Teams
15
Phase 2: Finding non-supported efficient teams
16
Experimental Results (Setup)
• Dataset:
• DBLP:
• Contains 5,658 experts and 8,588 edges.
• The expertise cost of an expert
• Number of publications of the expert.
• The more expertise the expert possesses, and thus the more expensive he/she is
17
WIC’14 Two-Phase Pareto Set Discovery for ...
Experimental Results (Setup)
• For each number of skills,
• 50 sets of skills are generated randomly
• Corresponding to 50 random projects.
• The average result for each number of skills is computed for
each algorithm.
• Performance Measure:
• Hypervolume, Average Distance, Maximal Distance, Precision, Recall,
Run Time
18
WIC’14 Two-Phase Pareto Set Discovery for ...
Results of the first phase
19
Results of the two phase method in comparison with other
methods on the DBLP graph.
20
RESULTS OF ALGORITHMS FOR FINDING
PARETO FRONT
21
Conclusion
• The problem of finding a team of experts in a social network
that minimizes the communication cost, expertise cost and
the personnel cost is defined.
• An approximation algorithm with a provable performance
bound is proposed to find pareto set.
• The algorithm consists of two main phases: finding supported
efficient teams and then uses heuristics to find none-
supported efficient teams.
22
WIC’14 Two-Phase Pareto Set Discovery for ...
WIC’14 Two-Phase Pareto Set Discovery for ...
Thank You

Two-Phase Pareto Set Discovery for Team Formation in Social Network

  • 1.
    Morteza Zihayat Mehdi Kargar AijunAn York University, Toronto, Canada Two-Phase Pareto Set Discovery for Team Formation in Social Network
  • 2.
    Overview • Team Formationin Social Networks • Communication Cost • Personnel Cost • Expertise level • Approximation Algorithm • Empirical Results • Conclusion 2 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 3.
    Team of Experts •Given a social network, find a team of experts that minimizes the communication cost, the personnel cost and the Expertise cost in order to complete a project. • Clearly, our problem is a three criteria optimization problem. 3 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 4.
    Team of Experts •Project: set of required skills • Expert: an individual with a specific skill-set • Social Network: represents strength of relationships (the degree of collaboration between any two experts). • For example: LinkedIn, DBLP and … 4 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 5.
    Team of Experts •Feasible Team of Experts: Given a set of experts and a project that requires a set of skills {s1, s2, . . . , sp}, a team of experts is a set of p skill-expert pairs: {(s1, cs1), (s2, cs2) , . . . , (sp, csp) }, where csk is an expert having skill sk for k = 1, . . . , p. • A skill-expert pair (sk, csk) • expert csk is responsible for skill sk in the project. 5 WIC’14 Two-Phase Pareto Set Discovery for ...  How to make sure that the experts can communicate together?  What about the personnel cost?  How to make sure that the experts are the best experts for the required skills (Expertise cost).
  • 6.
    Communication Cost • Anexperts social network is modeled as an undirected and weighted graph G. • For the following team of experts regarding graph G • The sum of distances of a team of experts • The sum of the shortest distances between the experts responsible for each pair of skills. 6 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 7.
    Personnel Cost • Forthe following team of experts regarding graph G • An expert c is paid k × t(c) • k is the number of skills the expert is responsible for in the project. • More skills the expert uses → More responsibility or tasks → More money he/she is received 7 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 8.
    Expertise Cost • Forthe following team of experts regarding graph G • Where is the expertise cost with respect to the skill and is reversely proportional to the expertise level of the expert. 8 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 9.
    Three Objective TeamFormation Problem • In three objective optimization problem: • There does not exist an answer that optimizes all the criteria simultaneously. • Instead of finding one single answer, we can find a set of teams that are not dominated by others. • Dominance Relation: Given two feasible teams T and T’, T dominates T’ if and only if CC(T)≤CC(T’), EC(T) ≤ EC(T’) and PC(T) ≤ PC(T’). It is denoted as • Efficient Team: Team T is efficient (also called Pareto- optimal) for a project P if and only if there does not exist a feasible team T’ for P such that 9
  • 10.
    Efficient Teams Categorization •The efficient teams can be categorized into two main types: • Supported Efficient Teams: can be obtained by solving the following weighted sum single-objective problem • We denote this weighted-sum single objective problem as WeightedSingleObjλ • Non-supported efficient team. An efficient team that is not an optimal team of any WeightedSingleObjλ problem. 10
  • 11.
    Pareto Front TeamFormation Problem • Given a project P and a graph G representing the network of experts, the Pareto front team formation problem is finding the set of all efficient teams from G for P. • We present a two-phase method for finding the set of efficient teams in graph for a project. • Lexicographic Optimal Team: Feasible team T is a lexicographic optimal team with respect to a lexicographic order of objectives (cost1,cost2,cost3) if for any other feasible team T’: • Cost1(T) < Cost1(T’) • Or, Cost1(T) = Cost1(T’) but Cost2(T) < Cost2(T’) • Or, Cost1(T) = Cost1(T’) and Cost2(T) = Cost2(T’) and Cost3(T) < Cost3(T’) 11
  • 12.
  • 13.
    Finding LOT whileOptimizing Communication Cost • Finding the optimal team while minimizing the communication cost is proved to be NP-hard • We find a LOT with respect to the lexicographic order of (CC,EC,PC) • Since minimizing the communication cost is NP-hard, this algorithm finds the best team with 2-approximation 13
  • 14.
    Finding LOT whileOptimizing Expertise Level • We propose a method for finding a LOT with the minimum expertise cost • This algorithm finds an exact LOT in polynomial time • Its time complexity is O(m2 × n) where m is the number of experts in G and n is number of required skills for project P. • The similar algorithm is applied for finding LOT while optimizing Personnel cost. The only difference is using personnel costs instead of expertise costs. 14
  • 15.
  • 16.
    Phase 2: Findingnon-supported efficient teams 16
  • 17.
    Experimental Results (Setup) •Dataset: • DBLP: • Contains 5,658 experts and 8,588 edges. • The expertise cost of an expert • Number of publications of the expert. • The more expertise the expert possesses, and thus the more expensive he/she is 17 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 18.
    Experimental Results (Setup) •For each number of skills, • 50 sets of skills are generated randomly • Corresponding to 50 random projects. • The average result for each number of skills is computed for each algorithm. • Performance Measure: • Hypervolume, Average Distance, Maximal Distance, Precision, Recall, Run Time 18 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 19.
    Results of thefirst phase 19
  • 20.
    Results of thetwo phase method in comparison with other methods on the DBLP graph. 20
  • 21.
    RESULTS OF ALGORITHMSFOR FINDING PARETO FRONT 21
  • 22.
    Conclusion • The problemof finding a team of experts in a social network that minimizes the communication cost, expertise cost and the personnel cost is defined. • An approximation algorithm with a provable performance bound is proposed to find pareto set. • The algorithm consists of two main phases: finding supported efficient teams and then uses heuristics to find none- supported efficient teams. 22 WIC’14 Two-Phase Pareto Set Discovery for ...
  • 23.
    WIC’14 Two-Phase ParetoSet Discovery for ... Thank You