关键词: 超关系知识图谱, 大语言模型, 超关系知识图谱补全, 超关系事实, 限定符

Abstract: Aiming at the problems of incomplete semantic representation in hyper-relational facts, insufficient task accuracy and generalization ability caused by the qualifier sparsity in Hyper-relational Knowledge Graph (HKG), a Large Language Model (LLM) driven method L-EQs (Large language model-driven method for Qualifier Enhancement in hyper-relational knowledge graphS) was proposed to alleviate qualifier sparsity and improve semantic completeness. First, semantically rich descriptions were extracted from knowledge bases to mitigate semantic ambiguity caused by identical labels of entities and relations. Then, prompt templates were used to guide the LLM to generate qualifiers and their semantic explanations as external knowledge. To ensure the quality of the qualifiers, an iterative prompting strategy was employed to filter the generated qualifiers, with multiple iterations conducted until the predetermined quantity was reached. Third, embeddings of external knowledge were constructed, and an external knowledge aggregation module was introduced to suppress noise during the aggregation process, thereby providing high-quality semantic support for the completion task. Finally, the task results were predicted through a decoder, and extensive experiments were conducted on the open-source datasets WikiPeople and WD50K to validate the effectiveness of the L-EQs method. Experimental results show that, compared to the optimal results of the baseline method, the MRR, Hit@1, Hit@5, and Hit@10 of L-EQs on the WikiPeople dataset increased by 8.4, 4.7, 5.3, and 6.8 percentage points, respectively. On the WD50K dataset, corresponding improvements of 4.8, 2.3, 2.7, and 3.0 percentage points were also observed. This significant improvement proves the effectiveness of the L-EQs in addressing the qualifier sparsity and alleviating the issues caused by the incomplete semantic representations.

Key words: Hyper-relational knowledge graph (HKG), Large language mode (LLM), hyper-relational knowledge graph completion, hyper-relational fact, qualifier