Information propagation prediction method and system based on cascading spatial-temporal characteristics

Abstract

The invention relates to an information propagation prediction method and system based on cascading spatial-temporal characteristics. The method comprises the steps: performing cascade conversion on given information into a vector form; capturing cascading time sequence features, and obtaining feature vectors of cascading in a time dimension; constructing a heterogeneous graph by using the social relation graph and the cascade graph, and capturing a dependency context of a network node; capturing topological characteristics of information spread in a network; performing fusion to obtain cascaded spatial-temporal features; according to the position relationship of the network nodes in the time dimension, weighting the historical spatial-temporal characteristics of the cascade at different moments to obtain weighted characteristic vectors of the cascade at different moments; employing multi-head self-attention for adjusting the importance of nodes at different cascading moments, and obtaining final cascading feature representation; performing information propagation prediction by using the final cascaded feature representation to obtain probability distribution of node activation at the next moment, and selecting the node with the maximum activation probability as the node activated at the next moment. According to the method, the node activated at the next moment can be better predicted.

Description

technical field [0001] The invention relates to social media-oriented information dissemination prediction, in particular to an information dissemination prediction method and system based on cascaded spatio-temporal features. Background technique [0002] Internet users generate massive amounts of information on social media platforms, and the spread of information on the network leaves a series of trajectories, thus forming a cascade. Information dissemination prediction is to use these cascades to learn the underlying rules and mechanisms of information dissemination, so as to predict the next affected user. It has a wide range of applications in marketing, public opinion supervision, election prediction and other fields. [0003] Information dissemination prediction technology has developed rapidly in recent years. At present, the relevant research on information dissemination can be mainly divided into three categories: (1) methods based on generative models, which us...

AI Technical Summary

Problems solved by technology

This method can explain well, but it is difficult to achieve good results in complex networks; (2) Based on the method of feature construction, the characteristics of user attributes, network structure, information content, etc. are constructed through manual rules, and then traditional machine learning is used to method to predict

This type of method requires a large human cost to carry out feature engineering, and its generalization ability is poor; (3) Based on deep learning methods, cascaded features are extracted in an end-to-end manner through recurrent neural networks or graph neural networks.

[0004] On the one hand, since the information cascade is usually expressed as a time-ordered sequence, the dependency information between some nodes is lost

On the other hand, information tends to spread more easily on user networks with social connections, thus forming cascaded spatio-temporal features. Existing methods usually use sequence modeling methods to extract cascaded context dependencies, but there is no method that can efficiently extract cascaded spatio-temporal features

Images