home..
lab meeting1
huyi / July 2022
2022/7/8 组会 (待完善)
小娟学姐 RulE:
Knowledge Graph 简介
KG completion:
- neural(KnowledgeGraphEmbedding)
- transE: 1t1
- TransR: multi-t-multi
- transE: 1t1
- symbolic reasoning
各有优缺点:
RulE
rule body–>rule head
modeling triples: RotateE
- 对KG进行embedding
aggregation:需要加入pair-wise部分?之后再加三条路……
rule跟entity相关?比如说“父亲的配偶是母亲”在不同年代的置信度是不一样的?把entity建模在前面一项 为什么后面要加KGE? 要把entity的信息加进去宇新学姐 ParameterAdaptive-GNN
idea:通过图的结构来调整parameter的规模
parameter network –> regularization $\Delta\gamma_{i,k}$ –> classification network
RNNlogic
naive的做法:将每个可能的rule作为一个candidate rule,然后score每个rule