pick_kols: Pick key opinion leaders from a network given constraints

In KOLaide: Pick and Plot Key Opinion Leaders from a Network Given Constraints

Pick key opinion leaders from a network given constraints

Description

Pick key opinion leaders from a network given constraints

Usage

pick_kols(
  network,
  tosource = TRUE,
  goal = "diffusion",
  m = 1,
  range = c(1, 1),
  top = NULL,
  include = NULL,
  exclude = NULL,
  attribute = NULL,
  alpha = 0.9,
  beta = 0.9,
  file = NULL
)

Arguments

network	a unipartite unweighted network as an adjacency matrix or igraph object
tosource	logical: edges point toward a source of information
goal	string: goal for the KOL team (either "diffusion" or "adoption")
m	integer: KOL team centrality parameter (m == 1 is equivalent to simple degree centrality)
range	vector: a vector of length 2 containing the minimum and maximum number of KOLs on a team
top	numeric: restrict scope to the top nodes with the highest degree, closeness, or betweenness (useful for large networks)
include	vector: names or indices of nodes that must be included on the KOL team
exclude	vector: names or indices of nodes that can not be included on the KOL team
attribute	string or vector: if network is an igraph object, the name of a node attribute. if network is an adjacency matrix, a vector containing a node attribute.
alpha	numeric: parameter to control relative weight of breadth and diversity in overall evaluation of KOL teams (0.5 \leq \alpha \leq 1)
beta	numeric: parameter to control weight of team size in overall evaluation of KOL teams (0 \leq \beta \leq 2)
file	string: filename to write a sorted list of possible KOL teams as a CSV.

Details

When seeking to diffuse a piece of information or encourage adoption of a behavior, it is often useful to recruit the assistance of key opinion leaders (KOL) in a network. pick_kols facilitates selecting members of a KOL team by returning a dataframe of possible teams. The selection of a KOL team often depends on several factors, which this function summarizes as ABCDE:

• Availability - The availability of individuals to serve as a KOL. This can be controlled by the include and exclude parameters.

• Breadth - The fraction of non-KOLs that the KOL team can influence. When goal=="diffusion", breadth is measured as the fraction of non-KOLs that a KOL team can reach in m steps (i.e., m-reach). When goal=="adoption", breadth is measured as the fraction of non-KOLs that are directly connected to at least m KOLs (i.e., m-contact).

• Cost - The number of KOLs to be recruited and trained (i.e., team size).

• Diversity - The fraction of values of attribute represented on the KOL team.

• Evaluation - Potential KOL teams must be compared and evaluated in a way that balances these considerations.

Evaluating KOL Teams

Potential KOL teams are evaluated on the basis of breadth (B), Cost (C), and (if attribute is provided), Diversity (D) using

\frac{B}{C^\beta} \mbox{ or } \frac{B^\alpha D^{1-\alpha}}{C^\beta}

The \alpha parameter can take values 0.5 < \alpha < 1 and controls the weight placed on breadth relative to diversity. Smaller values of \alpha place less weight on breadth and more weight on diversity, while larger values of \alpha place more weight on breadth and less weight on diversity. The default (\alpha = 0.9) places the majority of weight on the breadth of the network that KOL teams cover, while still considering the team's diversity (primarily as a tie-breaker).

The \beta parameter can take values 0 < \beta < 2 and controls the cost of larger KOL team members. Smaller values of \beta imply decreasing marginal costs, while larger values of \beta imply increasing marginal costs. The default (\beta = 0.9) assumes that team members have a slight diminishing marginal cost (i.e. the cost of each additional team member is slightly smaller than the previous one).

Interpreting Edge Direction

If network is a directed network, then tosource controls how the direction of edges is interpreted:

• tosource = TRUE (default) - An edge i -> j is interpreted as "i gets information from j" or "i is influenced by j" (i.e., the edge points toward a source of information or influence). This type of data usually results from asking respondents to nominate the people from whom they seek advice. In this case, actors with high in-degree like j are generally better KOLs.

• tosource = FALSE - An edge i -> j is interpreted as "i sends information to j" or "i influences j" (i.e., the edge points away from a source of information or influence). This type of data usually results from asking respondents to report the people to whom they give advice. In this case, actors with high out-degree like i are generally better KOLs.

Value

A sorted list containing a data frame of possible KOL teams with their characteristics, the network, m, goal, and (optionally) attribute

Examples

network <- igraph::sample_smallworld(1,26,2,.2)  #An example network
igraph::V(network)$name <- letters[1:26]  #Give the nodes names
igraph::V(network)$gender <- sample(c("M","F"),26,replace=TRUE)  #Give the nodes a "gender"
teams <- pick_kols(network,              #Find KOL teams in `network`
                   m = 2,
                   range = c(2,4),       #containing 2-4 members
                   attribute = "gender", #that are gender diverse
                   goal = "diffusion")   #and can help diffuse information
teams$teams[1:10,]  #Look at the top 10 teams

KOLaide documentation built on June 8, 2025, 1:16 p.m.