Few-Shot Knowledge Graph Completion Based on Selective Attention and the Transformer

To solve the problem of insufficient entity representation, this paper proposes a few-shot knowledge graph completion algorithm based on a selective attention mechanism and a transformer. First, in the selective encoder stage, selective attention is introduced to help the algorithm distinguish between important neighbors and noisy neighbors. Second, in the relation aggregator stage, the combined structure based on the transformer and LSTM neural network is adopted to encode and output the triplet relationship. Finally, in the matching processor phase, the representation of the reference set is aggregated and compared with the query set for similarity. The proposed model’s effectiveness and viability are validated through extensive experiments conducted on publicly available datasets. The results demonstrate the model’s ability to enhance knowledge graph completion in few-shot scenarios.