Opinion Dynamics and Social Group Information Efficiency: an agent-based model

Abstract

A platform for agent-based modelling of social processes is developed. The focus is on information exchange in a group of social agents. The dynamics of opinions on some topic is modelled, as well as the structure of social ties, which are weakened if opinions are distant and strengthened if opinions are close. In addition, to objectify the notion of "information efficiency of communication in a group", a test is proposed in which the structure of social ties is transferred to the structure of neural network connections, so that the strength of a tie is equated to a weight. Training of such a "sociomorphic" neural network was conducted and tested on the MNIST array, a test of the ability to recognisehandwritten digits from 0 to 9. The previously described effects of polarisation of opinions and division of the group into subgroups with high density of ties within subgroups and weak ties between agents from different subgroups were obtained in the variant of discussion of one topic. To approximate real groups, a topic selection algorithm was added to the basic opinion sharing algorithm, which is performed by comparing the cosine distance between semantic vectors - the vector of values assigned to each agent and the vector of each of the 9 topics. The topic vectors were generated based on the real news agenda of the news agency. The value vectors were taken from work in which a large sample of Russian youth was surveyed about their association with something most important in life. The modelling results show that the presence of several topics in the news agenda makes the communication process more flexible and multidimensional: even for the most semantically close to values and frequently chosen topics, polarisation of opinions was much slower than in the case of a single topic, and complete segmentation of the group was not observed over the modelled time period. This methodological approach can be useful for modelling conflicts in different communities, as well as for designing artificial communities with increased constructive creative potential. For example, the model can be used to optimise communications in complex multicultural societies