A news event prediction method based on cross-time step high-order correlation modeling

By constructing entity graphs and entity hypergraphs in news event prediction, and performing relation-time-aware message passing and hyperedge representation updates, the problem of pairwise and higher-order correlations across time steps is solved, enabling accurate prediction of future relationships.

CN122087131BActive Publication Date: 2026-07-24ZHEJIANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG UNIV
Filing Date
2026-04-23
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing news event prediction methods struggle to effectively capture pairwise and higher-order correlations between entities across time steps, especially addressing the mismatch between semantic representation and prediction objectives between dynamic graphs and temporal knowledge graphs.

Method used

By constructing an entity graph for relation-time-aware message passing, and utilizing an entity-hyperedge mapper and a message passing network between hyperedges, cross-temporal pairwise and higher-order correlations between entities are captured. An entity hypergraph is then constructed for message passing updates, and prediction is performed in conjunction with a convolutional knowledge graph module.

Benefits of technology

It effectively models complex interaction patterns between entities across time steps in news event prediction, enhances the semantic representation capability of edges, adapts to the relation prediction requirements of temporal knowledge graphs for news event prediction, and achieves accurate prediction of future relationships.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122087131B_ABST
    Figure CN122087131B_ABST
Patent Text Reader

Abstract

The application discloses a news event prediction method based on cross-time-step high-order correlation modeling, and belongs to the field of news event prediction. The method comprises the following steps: constructing an entity graph based on historical news events, performing relationship-time-aware message passing in the entity graph to obtain comprehensive entity representation, and capturing cross-time pair correlation between entities; passing the comprehensive entity representation through a learnable entity-superedge mapper to obtain an indication matrix describing the membership relationship between entities and superedges and construct an entity supergraph to capture the cross-time high-order correlation between entities; then, message passing is performed on the entity supergraph to update the superedge representation and further update the entity representation, which is used for predicting all possible relationships between entities in the future. The application effectively models the pair correlation and high-order correlation between entities in the news event prediction task across time steps, focuses on relationship semantic prediction, and adapts to the representation learning requirements of the relationship prediction-oriented representation learning in the news event prediction temporal knowledge graph.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • CN116894096A

  • CN119293538A