Research On User Personality Prediction Based On Social Network

Personality is a high summary of various personal characteristics among people,which is closely related to people’s psychological activities and behavior.Personality prediction aims to evaluate users’ personality by using their external performance and behavior characteristics.Personality recognition technology is of great significance in the fields of personalized service,medical service,employment and recruitment.In recent years,computer and Internet technology have risen rapidly,online social networking platforms such as Facebook,twitter,Myspace and Linked In are also developing rapidly,and the amount of information generated is growing exponentially.As people staying in these network platforms longer and longer,users have left a large number of digital traces rich in personality implicitly or obviously.Therefore,using these data to analyze,evaluate and predict users’ personality characteristics has attracted more and more attention of scholars because of their efficient and accurate characteristics.In the early days,the commonly used method in this field was to manually extract features from the above data,and use machine learning methods to predict personality through regression or classification models.However,in this process,the production of artificial features relies too much on expert experience.With the development of computer science fields such as image processing and artificial intelligence,more automatic personality recognition technologies have become common.Researchers began to avoid the use of artificial features through automatic supervised feature learning.Therefore,deep learning methods that can automatically mine deep-seated features,such as neural networks,have been further applied.However,the above methods rarely consider the relationship between users’ network structure and personality,which can play an important role in personality prediction.Moreover,in the cascade communication process,the communication behavior driven by users’ personality traits is often ignored.To solve the above problems,this dissertation studies the personality prediction algorithm based on graph neural network from two aspects: the user network structure and the user personality driving in cascade propagation.The main contribution of this work is as follows:1.Aiming at the lack of network structure characteristics in the existing personality prediction,a personality prediction algorithm based on Graph Neural Network(Deep Per)is proposed.The algorithm designs an end-to-end framework.Based on Graph Neural Network,the automatic recognition and calculation of user personality is carried out by using the attention network structure formed by users’ social interaction in the online virtual environment and its published text content.Firstly,the user’s local network is constructed,and the node embedding training is carried out on the network to obtain the pre training data of each node in the graph.Then,the learned node specific features are used as the input of graph neural network to learn the potential personality representation of users.Finally,the model outputs the big five personality score of each user node and compares it with the real value to minimize the log likelihood loss.The experimental results on real data sets prove the effectiveness of our proposed framework in predicting individual personality characteristics.2.Aiming at the lack of research on users’ personality driving force in the process of cascade propagation in social networks,a multi-task framework of information cascade and personality trait prediction based on Graph Neural Network is proposed(Me CAPE).The framework is a multi-task learning framework,which can carry out personality recognition task and cascade prediction task at the same time.Me CAPE constructs a two-layer Graph Neural Network,which is used to model cascade propagation and personality characteristics respectively.A gating mechanism,Personality Gate,is designed to simulate the interaction process between cascade communication and user personality.Experimental results on two real data sets demonstrate the effectiveness of our proposed framework in predicting cascades and predicting individual personality characteristics.