All Times EDT
Performance of a Genetic Algorithm for DeGroot Opinion Diffusion Model Parameter Estimation Under Assumption Violations (309939)Yuri A. Amirkhanian, Medical College of Wisconsin
Nicole Bohme Carnegie, Montana State University
*Kara Layne Johnson, Montana State University
Jennifer L. Walsh, Medical College of Wisconsin
Keywords: opinion diffusion, DeGroot model, genetic algorithm, social network intervention, pre-exposure prophylaxis (Prep), parameter estimation
An ongoing study leverages opinion diffusion to increase willingness to use pre-exposure prophylaxis (PrEP) for Black men who have sex with men. In order to assess the efficacy of the intervention using data on PrEP-related opinions of agents within the network, we use the DeGroot opinion diffusion model; however, existing algorithms for parameter estimation require large data sets that are often infeasible to obtain. In order to model this opinion diffusion process, we developed a novel genetic algorithm capable of recovering the parameters of a DeGroot model using small data sets. Our first simulation study assessed algorithm performance across a variety of networks and data set features: network size, degree, stubbornness of agents, number of time steps, and proportion of missing data. As these simulations do not reflect the issues present in many practical applications, we conduct a second simulation study assessing performance using ordinal data, incorrectly specified ties, agents missing from the network, and model misspecification. We present the results of the second simulation study along with parameter estimates from preliminary PrEP data in the context of these results.