.png){kind=icon}
.png){kind=icon}
Return
RLPath: a knowledge graph link prediction method using reinforcement learning based attentive relation path searching and representation learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-021-02672-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Due to containing rich patterns between entities, relation paths have been widely used in knowledge graph link prediction. The state-of-the-art link prediction methods considering relation paths obtain relation paths by reinforcement learning with an untrainable reward setting, and realize link prediction by path-ranking algorithm (PRA), which ignores information in entities. In this paper, we propose a new link prediction method RLPath to employ information in both relation paths and entities, which alternately trains a reinforcement learning model with a trainable reward setting to search high-quality relation paths, and a translation-based model to realize link prediction. Simultaneously, we propose a novel reward setting for the reinforcement learning model, which shares the parameters with the attention of the translation-based model, so that these parameters can not only measure the contributions of relation paths, but also guide agents to search relation paths that have high contributions for link prediction, forming mutual promotion. In experiments, we compare RLPath with the state-of-the-art link prediction methods. The results show that RLPath has competitive performance.
Keywords:
Knowledge graph link prediction
Representation learning
Reinforcement learning
Path searching
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W