基于嵌入特征和稀疏矩阵的实体对齐方法
|
|
冯超文,耿程晨,刘英莉
|
Entity alignment method based on embedding features and sparse matrices
|
|
Chaowen FENG,Chengchen GENG,Yingli LIU
|
|
| 表 3 所提实体对齐方法不同模块的消融实验结果 |
| Tab.3 Ablation results of different modules in proposed entity alignment method |
|
| 方法变体 | DBP_ZH-EN | | DBP_JA-EN | | DBP_FR-EN | | Hits@1 | Hits@10 | MRR | | Hits@1 | Hits@10 | MRR | | Hits@1 | Hits@10 | MRR | | 完整模型 | 0.871 | 0.950 | 0.900 | | 0.938 | 0.982 | 0.955 | | 0.976 | 0.995 | 0.984 | | 移除稀缺特征传播模块 | 0.836 | 0.932 | 0.878 | | 0.902 | 0.971 | 0.930 | | 0.942 | 0.985 | 0.962 | | 移除关系嵌入模块 | 0.819 | 0.924 | 0.867 | | 0.889 | 0.958 | 0.923 | | 0.934 | 0.982 | 0.958 | | 移除字符级嵌入模块 | 0.850 | 0.940 | 0.890 | | 0.921 | 0.975 | 0.942 | | 0.960 | 0.990 | 0.975 | | 移除Sinkhorn 归一化模块 | 0.845 | 0.937 | 0.885 | | 0.918 | 0.973 | 0.940 | | 0.956 | 0.988 | 0.973 | | 移除Hungarian算法模块 | 0.790 | 0.910 | 0.840 | | 0.864 | 0.946 | 0.905 | | 0.918 | 0.975 | 0.948 |
|
|
|