A Contextual Information-Augmented Probabilistic Case-Based Reasoning Model for Knowledge Graph Reasoning

Knowledge Graph Reasoning (KGR) is one effective method to improve incompleteness and sparsity problems, which infers new knowledge based on existing knowledge. Although the probabilistic case-based reasoning (CBR) model can predict attributes for an entity and outperform other rule-based and embedding-based methods by gathering reasoning paths from similar entities in KG, it still suffers from some problems such as insufficient graph feature acquisition and omission of contextual relation information. This paper proposes a contextual information-augmented probabilistic CBR model for KGR, namely CICBR. The proposed model frame the reasoning task as the query answering and evaluates the likelihood that a path is valuable at answering a query about the given entity and relation by designing a joint contextual information-obtaining algorithm with entity and relation features. What's more, to obtain a more fine-grained representation of entity features and relation features, the CICBR introduces Graph Transformer for KG's representation and learning. Extensive experimental results on various benchmarks prominently demonstrate that the proposed CICBR model can obtain the state-of-the-art results of current CBR-based methods.