Graph Neural Networks in Analysis of Main Influencing Factors and Modeling of Network Content Propagation Rule

EasyChair Preprint 9315

Abstract

In recent years, various social media platforms have come into being, and the information diffusion through social networks has become more frequent and complex. Research on the rules for dissemination of information and its influencial factors have become significant in field of networks. Further, it is necessary to regulate harmful information and also provide recommendations. Based on the graph neural network, this paper proposes an ICGNN model, studies the factors that affect information diffusion, and predicts whether a  blog will eventually be forwarded to other users (message receiver) by the message sender. It also analyses the factors that affect the rule of message transmission, and conducts a comparison of these factors with case analysis. During the whole experiments, firstly, this paper uses the public microblog data set, and collects and analyzes user characteristics, network structure factors and information content affecting information diffusion. Secondly, Jieba word segmentation and Bert are used to process text features, and node features are obtained by filtering incomplete nodes, achieving graph embedding. Additionally, the independent cascading model is reproduced and trained to converge. Finally, a multi-label classification model was designed. After that, features were set into a graph convolutional network and graph attentional network for comparative experiments, and then labels were restored by threshold for model evaluation. Through the verification and the testing of the model, the recall rate is 52% and the accuracy rate is up to 69%. The results show that the ICGNN model has good ability to predict the pattern of information diffusion. Moreover, a prediction result  is selected for case analysis, compare the information and influencing factors of the message sender and receiver, summarize the influence degree of some influencing factors, and put forward future research direction.

Keyphrases: Graph Neural Network, Natural Language Processing, independent cascade model, information dissemination, multi-label classification, social networking

@booklet{EasyChair:9315,
  author    = {Yupei Li},
  title     = {Graph Neural Networks in Analysis of Main Influencing Factors and Modeling of Network Content Propagation Rule},
  howpublished = {EasyChair Preprint 9315},
  year      = {EasyChair, 2022}}