Preference elicitation in synthetic social networks
The paper considers a scenario in which public administration (PA) uses an online social platform to collect information on citizens' preferences. However, the opinions of the sub-population that uses the online platform might be not representative. The author develops a method for generalization of the dynamics of the preferences observed on the social platform onto the entire population. The available data include information collected by the PA from the online platform (assuming that it is run and administered by the PA) and census data regarding the population. Hence, the PA has access to basic personal data of platform users (e.g. gender and age), position in the online social network, and opinions revealed on the platform. The online users' data can be analyzed along with the aggregated census data on the entire population. The author has implemented a multi-agent simulation model that takes into account the distribution of personal attributes, social network data, and opinion diffusion dynamics. The analysis involves showing how different algorithms enable generalization of preferences collected by the online platform to the entire population. The results of the analysis prove that the proposed method is efficient in the preference elicitation process – with each simulation step, the preference congruence level between real and synthetic populations increases. The main determinants of preference elicitation errors include the preference diffusion model and the weight of the agents’ own opinions.
The author has read and accepts the terms contained in the license agreement