Entity Alignment Model Based on Multi-Hop Information Fusion
Entity alignment is a critical step in knowledge graph fusion from various sources.Existing entity alignment methods primarily take advantage of structural information and entity names whilst ignoring attribute information in most cases.In terms of structure utilization,it primarily utilizes the structure of a first-order neighborhood for information transmission and lacks the perception of distant neighbors.To address these issues,an entity alignment model based on multi-hop information fusion is proposed herein.A pre-trained language model is used to encode the attribute value information.The entity name and attribute values are input into different model encoders to achieve information fusion.The attention mechanism is used to fuse the entity information at different distances.Distance matrices under different information representations are calculated,and the final alignment result is obtained after the matrix is fused and adjusted.Based on experiments with the original and degraded DBP15K datasets,it can be observed that the proposed model achieves more accurate alignment results overall,compared with the baseline models,where the Hits@1 performance is increased by up to 2.51 and 5.54 percentage points over state-of-the-art models.
entity alignmentknowledge graphGraph Neural Net work(GNN)attention mechanismknowledge fusion
万方数据
维普
