Attraction-Repulsion Weighted Hegselmann-Krause

The Attraction-Repulsion Weighted Hegselmann-Krause was introduced by Toccaceli et al. in 2020 [1].

This model is a variation of the Weighted Hegselmann-Krause (WHK). This model considers pair-wise interactions. To model the attraction and repulsion of opinions, during each iteration an agent \(i\) is randomly selected along with one of its neighbors, \(j\) - not taking into account the \(\epsilon\) threshold. Once identified the pair-wise interaction, the absolute value of the difference between the opinions of \(i\) and \(j\) is computed. There are four different variants of the method:

  1. Base case: If the computed difference value is lower than \(\epsilon\) then the update rule becomes:
\[\begin{split}x_i(t+1)= \left\{ \begin{array}{ll} x_i(t) + \frac{x_i(t) + x_j(t)w_{i,j}}{2} (1-x_i(t)) & \quad \quad \mbox{if } x_i(t) \geq 0\\ x_i(t) + \frac{x_i(t) + x_j(t)w_{i,j}}{2} (1+x_i(t)) & \quad \quad \mbox{if } x_i(t) < 0 \end{array} \right.\end{split}\]
  1. Attraction: if the computed difference value is lower than \(\epsilon\) then the following update rule are applied:
\[\begin{split}x_i(t+1)=\begin{cases} x_i(t) - \frac{sum_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) \geq 0, x_i(t) > x_j(t) }\\ x_i(t) + \frac{sum_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) \geq 0, x_i(t) < x_j(t) } \\ x_i(t) + \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) < 0, x_i(t) > x_j(t) }\\ x_i(t) - \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) < 0, x_i(t) < x_j(t) } \\ x_i(t) - \frac{sum_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) < 0, sum_{op} > 0}\\ x_i(t) + \frac{sum,_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) < 0, sum_{op} < 0}\\ x_i(t) + \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) \geq 0, sum_{op} > 0}\\ x_i(t) - \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) \geq 0, sum_{op} < 0}\\ \end{cases}\end{split}\]

where \(sum_{op} = x_i(t) + x_j(t)w_{i,j}\).

  1. Repulsion: if the difference between \(x_i(t)\) and \(x_j(t)\) exceeds \(\epsilon\) then the following update rule are applied:
\[\begin{split}x_i(t+1)=\begin{cases} x_i(t) + \frac{sum{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) \geq 0, x_i(t) > x_j(t) }\\ x_i(t) - \frac{sum_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) \geq 0, x_i(t) < x_j(t)} \\ x_i(t) - \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) < 0, x_i(t) > x_j(t) }\\ x_i(t) + \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) < 0, x_i(t) < x_j(t) } \\ x_i(t) + \frac{sum_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) < 0, sum_{op} > 0}\\ x_i(t) - \frac{sum_{op}}{2} (1-x_i(t)) & \text{if x_i(t) \geq 0, x_j(t) < 0, sum_{op} < 0}\\ x_i(t) - \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) \geq 0, sum_{op} > 0}\\ x_i(t) + \frac{sum_{op}}{2} (1+x_i(t)) & \text{if x_i(t) < 0, x_j(t) \geq 0, sum_{op} < 0}\\ \end{cases}\end{split}\]

where \(sum_{op} = x_i(t) + x_j(t)w_{i,j}\).

4. Attraction and Repulsion: if the computed difference value is lower than \(\epsilon\) then the attraction interaction occurs, otherwise the repulsion attraction occurs.

Statuses

Node statuses are continuous values in [-1,1].

Parameters

Name Type Value Type Default Mandatory Description
epsilon Model float in [0, 1] — True Bounded confidence threshold
perc_stubborness Model float in [0, 1] 0 False Percentage of stubborn agent
similarity Model int in {0, 1} 0 False The method use the feature of the nodes ot not
option_for_stubbornness Model int in {-1,0, 1} 0 False Define distribution of stubborns
method_variant Model int in {0, 1, 2, 3} 0 False The choice of the method to apply
weight Edge float in [0, 1] 0.1 False Edge weight
stubborn Node int in {0, 1} 0 False The agent is stubborn or not
vector Node Vector of float in [0, 1] [] False Vector represents the character of the node

Methods

The following class methods are made available to configure, describe and execute the simulation:

Configure

class ndlib.models.opinions.ARWHKModel.ARWHKModel(graph)

Model Parameters to be specified via ModelConfig :param epsilon: bounded confidence threshold from the HK model (float in [0,1]) :param perc_stubborness: Percentage of stubborn agent (float in [0,1], default 0) :param option_for_stubbornness: Define distribution of stubborns (in {-1, 0, 1}, default 0) :param similarity: the method uses the similarity or not ( in {0,1}, default 0) :param weight: the weight of edges (float in [0,1]) :param stubborn: The agent is stubborn or not ( in {0,1}, default 0) :param vector: represents the character of the node (list in [0,1], default []) :param method_variant: the variant of method to apply: 0-> base case 1->with attraction 2->with repulsion, 3-> with attractiona nd repulsion ( in {0,1, 2, 3}, default 0)

ARWHKModel.__init__(graph)

Model Constructor :param graph: A networkx graph object

ARWHKModel.set_initial_status(self, configuration)

Override behaviour of methods in class DiffusionModel. Overwrites initial status using random real values.

ARWHKModel.reset(self)

Reset the simulation setting the actual status to the initial configuration.

Describe

ARWHKModel.get_info(self)

Describes the current model parameters (nodes, edges, status)

Returns:a dictionary containing for each parameter class the values specified during model configuration
ARWHKModel.get_status_map(self)

Specify the statuses allowed by the model and their numeric code

Returns:a dictionary (status->code)

Execute Simulation

ARWHKModel.iteration(self)

Execute a single model iteration :return: Iteration_id, Incremental node status (dictionary code -> status)

ARWHKModel.iteration_bunch(self, bunch_size)

Execute a bunch of model iterations

Parameters:
  • bunch_size – the number of iterations to execute
  • node_status – if the incremental node status has to be returned.
  • progress_bar – whether to display a progress bar, default False
Returns:

a list containing for each iteration a dictionary {“iteration”: iteration_id, “status”: dictionary_node_to_status}

Example

In the code below is shown an example of instantiation and execution of an ARWHK model simulation on a random graph: we assign an epsilon value of 0.32, the percentage of stubborness equal 0.2, the distribution of stubborness equal 0 and a weight equal 0.2 to all the edges.

import networkx as nx
import ndlib.models.ModelConfig as mc
import ndlib.models.opinions as opn

# Network topology
g = nx.erdos_renyi_graph(1000, 0.1)

# Model selection
model = opn.ARWHKModel(g)

# Model Configuration
config = mc.Configuration()
config.add_model_parameter("epsilon", 0.32)
config.add_model_parameter("perc_stubborness", 0.2)
config.add_model_parameter("option_for_stubbornness", 0)
config.add_model_parameter("method_variant", 2)

# Setting the edge parameters
weight = 0.2
if isinstance(g, nx.Graph):
    edges = g.edges
else:
    edges = [(g.vs[e.tuple[0]]['name'], g.vs[e.tuple[1]]['name']) for e in g.es]

for e in edges:
    config.add_edge_configuration("weight", e, weight)


model.set_initial_status(config)

# Simulation execution
iterations = model.iteration_bunch(20)
[1]
  1. Toccaceli, L. Milli and G. Rossetti. “Opinion Dynamic modeling of Fake News Perception,” in Proceedings of International Conference on Complex Networks and Their Applications, 2020.