Query-Specific Context-Enhanced Representation Learning for Temporal Knowledge Graph Reasoning

Le résumé fourni par la source

Temporal Knowledge Graph Reasoning (TKGR) aims to leverage historical information to predict future facts. However, most existing methods learn entity embeddings shared across all queries at the same timestamp, without considering query-specific context. To tackle these problems, we propose Query-Specific Context-Enhanced Representation Learning (QSCEL) model for TKGR. Specifically, we construct an extended graph based on the query set to provide a structural basis for learning query-specific context. On this graph, we design a query-aware dual-attention graph neural network to learn both shared and query-specific contexts. Moreover, we design a global repeated history encoder that encodes fact frequency and temporal decay to better exploit historical information. Extensive experiments on four public datasets demonstrate the effectiveness of QSCEL. Our implementation is available at https://github.com/yugao9892/QSCEL.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.