.png){kind=icon}
.png){kind=icon}
Return
Sparse temporal knowledge graph completion based on path imitation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.129473.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Sparse temporal knowledge graph completion is a particular task in temporal knowledge graph completion, which aims to use the sparse knowledge in the sparse temporal knowledge graph to complete the missing facts. How to reduce the influence of sparsity in sparse temporal knowledge graphs is the critical issue to improve the completion effect. Existing completion models for sparse temporal knowledge graphs reduce the sparsity of temporal knowledge graphs by introducing additional action space and adaptive rewards, but these models still have limitations. On the one hand, the inference process of these models does not take the influence of noise information into account. On the other hand, the one-sidedness of the reasoning path of model training is also an important factor restricting the completion effect. In order to solve the above problems, we propose a Sparse Temporal knowledge graph completion model based on Path Imitation (STPI). In order to solve the noise problem, we propose a dynamic completion strategy based on path imitation. To solve the problem of one-sided paths, we quantify the similarity of paths and introduce them into soft rewards for reward shaping. Taking into account the limited data present in the sparse temporal knowledge graph, we gather details from neighboring entities to create updated entity embeddings. This method aids in enhancing the precision of our inference outcomes. Experimental results show that STPI outperforms state-of-the-art models on sparse public datasets while ensuring interpretability.
Keywords:
Temporal knowledge graph completion
Multi-reward
Path imitation
Reinforcement learning
Sparse temporal knowledge graph