基于关系平滑去噪与神经过程的小样本知识图谱补全方法OA

In order to solve the issues of noise interference and inadequate OOD generalization in FKGC,this proposed a relation-aware smoothing denoising neural processes framework,consisting of two modules:relation-aware smoothing denoising and uncertainty-aware generalization modeling.Firstly,it applied semantic smoothing to reduce bias caused by incorrect labels,and pruned false connections based on edge confidence to optimize the structure.To enhance generalization on complex patterns and anomalous data,it used a neural processes combined with normalized flows and stochastic decoders to model and strengthen the functional distribution of relations.Experiments conducted on three public benchmark datasets,NELL,WIKI,and FB15K-237.It demonstrates that the proposed algorithm outperforms ten other algorithms,achieving significant improve-ments of 7.1,4.1,and 5.1 percentage points in MRR,and 1.1,3.2,and 6.7 percentage points in hits@10,respectively.The experimental results confirm that the proposed framework achieves higher prediction accuracy and stronger generalization capability,validating its effectiveness in FKGC.