In Search Of Diverse And Connected Teams: A Computational Approach To Assemble Diverse Teams Based On Members Part 3

Number of objective functions

The third dimension is the number of objectives being optimized by the team formation algorithm. Some examples are minimizing teams' communication costs, minimizing teams' personnel costs, and maximizing the number of skills present in each team. 

The relationship between team formation algorithms and memory is closely linked. A team is a group of people, each with their ideas and abilities, but greater value can only be achieved when everyone works together.

The core of the team formation algorithm is how to make different people work together more harmoniously. In this process, everyone needs to use their strengths according to their roles and tasks, and at the same time need to communicate and coordinate effectively with other members of the team.

Memory plays an important role in this process. In a team, it is necessary to continuously record each member's tasks and contributions, as well as the team's progress and problems. Only in this way can effective communication and collaboration be formed in the team, and it can also help team members better understand their responsibilities and roles.

Furthermore, team formation algorithms and memory can also reinforce each other. Team formation algorithms can help people better understand how to work together, and developing stronger memories in the process will also allow people to better record and understand various information about the team.

Therefore, we should recognize the importance of team formation algorithms and memory for a team. Only through continuous communication and collaboration, as well as recording and organizing information, can the team operate more efficiently and exert greater value. It can be seen that we need to improve memory, and Cistanche deserticola can significantly improve memory, because Cistanche deserticola can also regulate the balance of neurotransmitters, such as increasing the levels of acetylcholine and growth factors. These substances are very important for memory and learning. In addition, Meat can also improve blood flow and promote oxygen delivery, which can ensure that the brain receives sufficient nutrients and energy, thereby improving brain vitality and endurance.

Click Know to improve short-term memory

Most algorithms define the team formation problem with a single objective with restrictions [59]. 

The examples mentioned before follow this single-objective function design. A pitfall is that other beneficial goals for team composition cannot be considered during the optimization process simultaneously (e.g., minimizing communication costs while maximizing the team's skills).

Prior studies have introduced more than one objective function to the team formation problem. One example is Kargar et al. [60], which presents the "Minimal Cost Contribution" algorithm (MCC). Its goal is to search for the team with the lowest communication costs and the lowest personnel costs simultaneously. 

MMC's objective function is a linear combination of both cost functions with a parameter λ that indicates the trade-off between communication and personal costs. This algorithm implements a heuristic approach that adds new members to the team incrementally and considers the costs of adding a new member concerning the current costs of the assembled team. 

Despite the benefits of these linear combination formulations, this approach presents two limitations: it provides only one single team solution, and its trade-off variable for the cost functions must be set in advance. Thus, finding other suitable solutions using these methods depends on the adjustment of the trade-off variable, which can add bias to the search process [61].

Recent algorithmic contributions have formulated the team formation problem as a multiobjective optimization problem to optimize two or more objective functions simultaneously [62, 63]. 

These problems involve trade-offs between two or more objectives since improving a solution in one objective is possible only by conceding another objective. Thus, multi-objective optimization problems do not provide a single solution but obtain multiple solutions considering different relevance emphases for the multiple objectives. 

While in single-objective optimization problems, the superiority of one solution over others is determined by the objective function, in multi-objective optimization problems it is determined by dominance. The optimization process looks for solutions that are better than others in all the objective functions. 

As a result, the problem delivers a set of "non-dominated" solutions, which consists of solutions that can be improved without a simultaneous detriment to at least one of the other objectives. Multi-objective optimization is also known as Pareto optimization. 

Fig 1 shows an example of a Pareto front showing different non-dominated solutions ranging between two objectives. Computing this Pareto front allows decision-makers to compare and check different trade-offs among both dimensions.

Based on this approach, multi-objective algorithmic implementations provide a set of team solutions that consider varied assessments of the objective functions [54, 64]. Zhang and Zhang's implementation [64] selects the members with the highest capabilities for the task and the best interpersonal relationships to assemble the best team. This study uses the particle swarm optimization implementation to determine whether a member must be part of the best team. 

Solutions move in a two-dimensional continuous space, and the algorithm applies a sigmoid function to binarize members' presence. Perez-Toledano et al. [63] developed a genetic algorithm to find competitive basketball teams considering the cost and valuation of each player simultaneously. 

Each solution consists of a team from a set of available players, and its final Pareto front displays different teams that consider the trade-off between players' valuation and cost. Based on these formulations, team builders can see and compare other teams and choose what objective they will prioritize when selecting a team.

Problem formulation

After reviewing relevant team formation problems and their respective algorithms, we aim to implement this particular problem that maximizes teams' diversity and teams' familiarity simultaneously. 

This problem is appropriate for a multi-objective optimization formulation since maximizing teams' familiarity could lead to forming groups with members that are similar to each other [65].

Although we could implement this problem as a single-objective optimization problem, we would have to prioritize one of those goals and avoid trade-offs among solutions. Moreover, prior formulations of the team formation searched for either the best team among multiple objectives or team combinations based on a single objective.
We propose a multi-objective optimization problem that assigns all available individuals into teams, resulting in several team combinations that consider different relevance emphases for diversity and familiarity. This work is not the case for previous studies on team formation and provides a new approach to the team formation literature.

Materials and methods

In this section, we introduce the multi-objective problem and definitions that we will use throughout this paper. Our notation is also summarized in Table 1. We also describe the NSGA-II implementation of this multi-objective problem and its components. We then describe the datasets and benchmark algorithms we used to evaluate the team formation problem. Finally, we explain the quantitative metrics to compare algorithms' results.

Definitions

Members, attributes, networks, and teams. We consider a set of participants P = {p1, p2, . . ., pn} with a set of categorical attributes C = {c1, c2, . . ., cm} and a set of numerical attributes U = {u1, u2, . . ., ul }. 

These individuals' attributes have different scales and represent information about each person (e.g., age, gender, race, skill). Depending on the individual information available, teams can have several attributes describing their qualities and composition. Each person has a value in each one of these attributes. We denote ci(pj) to obtain the value of the categorical attribute ci for the person j. 

Similarly, we use ui(pj) to obtain the value of the numerical attribute ui for the person j. Person j can be represented as a vector of these categorical and numerical attributes. Thus, we have the attributes of pj as (c1(pj), . . ., cm(pj), u1(pj), . . ., ul (pj)).

People are connected in a social network modeled as an undirected and unweighted graph G. We define G = (P, E), where E represents the graph's edges. Each node in G represents a person from P. We use person and node interchangeably throughout this paper. Two people are connected by an edge if they have collaborated in the past. In other words, if individuals i and j have worked together, then Gi,j = 1. Otherwise, Gi,j = 0.

Given this list of participants P connected in the network G, the goal is finding a set of teams T = {t1, t2, t3, . . ., tq}, where all members of P assemble q teams and belong to one team only. The optimization dual problem can be formulated as minimizing the communication costs among team members and maximizing teams' diversity levels. We now make these notions and describe each objective function.

Communication costs. Lappas et al. [57] focused on the importance of collaborations and familiarity between experts by considering the cost of their collaborations. According to this model, experts who collaborated in the past are more likely to exchange information and ideas effectively than experts without prior collaborations. 

Based on experts' prior collaborations, this model calculates the communication costs among team members to estimate their collaboration and familiarity levels. The goal of optimizing communication costs is to form teams with high familiarity levels. A literature review shows that communication costs are a highly used proxy for collaboration and familiarity among researchers [66].

In our setting, we use communication costs as a proxy for teams' familiarity. Kargar and An [31] found the total sum of distances between team members to be a reasonable measure of communication costs, as it is more stable to changes in the network than other potential measures. 

Other alternatives for communication costs are the social network's diameter (i.e., the largest shortest path between any two nodes in the network), and the minimum spanning tree (i.e., the minimum sum of the weights of a network's edges) [57]. 

We also implemented this problem using these two definitions, and their results were similar to those obtained using the sum of distances. The results of the diameter's implementation are available in S1 Fig and S1 Table in S1 File, and the results of the minimum spanning tree's implementation are available in S2 Fig and S2 Table in S1 File.

We define the communication costs between two individuals pi and pj, denoted as d(pi, pj), as the shortest path length while traversing the edges of the graph G from one node to another. If Pi and PJ have collaborated in the past, they are at a one-hop distance. 

If Pi and PJ have not collaborated but have a prior collaborator in common, they are separated by two shops. Having common past collaborators within a team can promote familiarity based on "triadic closure" [67]. 

This mechanism posits that nodes are more likely to establish a new connection when they have a connection in common. Three-hops and 4-hops can follow the same principles based on "balance mechanisms" [67]. 

Individuals will tend to forge new connections with collaborators of their collaborators to seek consistency within their group. Therefore, using the total sum of distances in our objective function aims to search for teams that maximize the number of direct collaborations (i.e., one-hops), common connections (two-hops), and close connections (three-hops or higher). 

The lowest communication cost value is when all team members have collaborated (i.e., they are directly connected), and the highest is when team members are not connected at all. In this implementation, if there is no path between pi and pj in G, we set the communication costs between them as the social network's diameter.
We define the communication costs of a team t as the total sum of shortest path lengths between members, as it is more stable to changes in the network than other potential measures. We denote by Cc(t) the communication costs of team t, which has k members. Thus, we define the team's communication costs as:

Cct ¼ Xk i;j2t;i6¼j dðpi ; pj Þ ð1Þ

The goal is to minimize the average sum of shortest path lengths across all assembled teams in the individuals' network. Computing the sum of communication costs of a set of teams runs in O(n2 ) time.

Team diversity score. The second goal is to generate diverse teams with a broad array of backgrounds, traits, and skill repertoires. Diversity describes the distribution of differences among the members of a unit concerning a common attribute [30]. 

Harrison and Klein [30] presented a framework suggesting that diversity is best conceptualized in three ways: separation, variety, and disparity. Separation refers to differences among team members in their lateral position on a continuum (e.g., value, attitude, belief). Variety refers to categorical differences among team members wherein the number of represented categories contribute to team diversity (e.g., gender, career, race). 

Finally, disparity represents differences in the concentration of valued assets or desirable resources (e.g., expertise, educational level, tenure). These metrics allow researchers to operationalize functional and demographic diversity in parallel and according to their theoretical conceptualizations [14].

For more information:1950477648nn@gmail.com