A Study of Internet Users’ Heterogeneity Based on Large Language Models

The emergence of big events is frequently followed by the existence of online public opinion, and netizens’ emotional comments heighten the hazards associated with the process of online public opinion transmission, resulting in social instability. Netizens’ emotionality is inextricably linked to their personality traits, and it is necessary to analyze their personalities in order to determine the corresponding public opinion countermeasures of government departments; thus, this paper conducts a study on netizen heterogeneity. A Graph Neural Network (GNN) personality prediction model is built on the Big Five personality model and the BERT model. The BERT model encodes the comment data and converts it into a graph neural network model; the dual attention mechanism extracts the relationship between the user and his/her diction, diction, and psychological keywords; the graph convolutional neural network extracts the features; and finally, the personality scores of the five dimensions of the Big Five personality model of Internet users are calculated. The study demonstrates that the graph neural network Internet user personality prediction model based on the BERT model has some advantages in the personality prediction task and can give data support for future Internet public opinion analysis.