Relation Aware Neural-Symbolic Models for Logical Queries on Knowledge Graphs

Complex multi-hop logical queries are a highly regarded problem in the field of knowledge graphs. The challenge lies in whether the query model can effectively rely on observed entities and edges to handle incomplete knowledge graphs in the real world. This requires the model to be able to correctly retrieve answers even when encountering unobserved intermediate entities or relations while processing complex logical queries involving missing information. In this paper, we propose a neural-symbolic integration approach to address the first-order logical query problem on knowledge graphs and make improvements and optimizations to this approach. The main contributions of this paper are summarized as follows: (1) We design and implement a neural-symbolic integration model for knowledge graph logical queries, which consists of symbol query decomposition, neural relation projection, and fuzzy logical inference components. (2) We propose a relation-aware model for complex logical queries on knowledge graphs, which includes a newly designed graph neural network with residual structures and a relation-enhancement module to improve the accuracy and generalization capability of the model. Experimental results on the FB15k-237 dataset validate the effectiveness of the proposed model for knowledge graph logical query tasks. Finally, we summarize the achievements of this paper and provide further research prospects.