|
|
The search for gathering a team of experts who are expected to collaboratively work towards accomplishing a given project successfully is referred to as Team Formation, a problem that has historically been solved in a variety of ways including manually in a time-consuming and bias-filled manner, and, algorithmically within such disciplines as social sciences, management, operations technology, and so forth. In the present effort, while providing a taxonomy to distinguish between search-based versus learning-based approaches, we survey the graph-based Team Formation studies from the search-based category, motivated as they comprise the mainstream. We present a unifying and vetted overview of the various definitions of the notions in this realm, scrutinize assumptions and benchmarks, and identify shortfalls. We start by drawing upon initial efforts and approaches to the problem of Team Formation to lay the conceptual foundations and set forth the necessary notions for a more grounded view of this realm. Next, with the help of a specifically designed set of notations, we provide a detailed view of the graph-based Team Formation approaches based on the objective functions they aim to optimize. In doing so, we lay out who builds on whom and how algorithms have evolved to solve previous works’ drawbacks. Additionally, we categorize different evaluation schemes and elaborate on different metrics and the insights that can be drawn from each. Referring to the evaluation schemas and metrics, we compare different works and propose future directions.
Our survey brings forth a unifying and vetted methodology to the various definitions of the notions in this realm, criticizes assumptions and comparative benchmarks, and points out shortfalls to smooth the path for future research directions. In this survey, we present a novel taxonomy from a computational perspective as shown above.
We present a comprehensive overview of 17 seminal graph-based works on the Team Formation problem,
including 13 proposed optimization objectives, after screening 63 algorithms from 126 papers.
We also recognize the significance of learning-based and operations research (OR)-based methods for Team Formation. We have targeted them in our immediate research direction.
The search for an almost surely successful team (optimum team) can be traditionaly formulated as a variant of the set cover problem on subsets of experts, using operations research (OR) techniques such as integer linear and nonlinear programming. However, by expanding graph-based methods—which view a team as inherently relational and focus on the interactions and collaboration among experts—we observe a synergistic integration of expert graphs with OR techniques. This integration allows the modeling of the problem through expert graph concepts while leveraging OR methods to solve it effectively.
─categirized_papers
├───capacity_of_team_members
├───constraint
│ ├───authority
│ ├───communication_cost
│ │ ├───buttleneck
│ │ ├───dense
│ │ ├───diameter
│ │ ├───graph_clustering
│ │ ├───stainertree
│ │ └───sum_of_edge_weight
│ ├───geographical proximity
│ ├───trust
│ └───workload
├───dynamic_network
├───efficiency
├───fairness
├───grouped_team
├───keyword_search_and_community_search
├───learning_based
│ ├───game_theory
│ └───learning_search_based
├───multi_objectiver
├───number_of_created_teams
│ ├───more_than _one
│ │ ├───pareto_set
│ │ └───top_k
│ └───one
├───old
│ ├───education
│ ├───engineering
│ ├───multi_skill_heuristic_solution
│ ├───network
│ ├───performance
│ └───team_member_characteristic
├───operation_research
│ ├───fuzzy
│ ├───genetic_algorithm
│ ├───hierarchical
│ ├───integer_programming
│ └───linear_programming
├───similarity_between_two_graphs
│ ├───graph_pattern
│ └───kernel_replacing_a_member
├───surveys
├───team with leader
└───team_size
├───at_least_k_person_for_each_skill
├───at_most_k_responsiblity_for_each_person
└───small_teams