🛣[Deep Learning]Stanford CS224w:Machine Learning with Graphs
想说的话🎇
🔝课程网站：http://web.stanford.edu/class/cs224w/
👀一些资源: B站精讲：https://www.bilibili.com/video/BV1pR4y1S7GA/?spm_id_from=333.337.search-card.all.click&vd_source=280e4970f2995a05fdeab972a42bfdd0
https://github.com/TommyZihao/zihao_course/tree/main/CS224W
Slides: http://web.stanford.edu/class/cs224w/slides
Knowledge Graphs
TransE
For a triple \((h,r,t)\),let \(h,r,t \in \mathbb{R}^d\) be embedding vectors.
TransE: \(h + r ≈ t\) if the given link exists else \(h + r ≠ t\).
Entity scoring func:
对比损失(Contrastive loss)：对有效的三元组支持较低的距离（或较高的分数），对损坏的三元组则支持较高的距离（或者较低的分数）
Connectivity Patterns in KG
-
Symmetry: If the edge \((h,"Roommate",t)\) exists in KG, then the edge \((t,"Roommate",h)\) should also exist.
-
Inverse relation : If the edge \((h,"Advisor",t)\) exists in KG, then the edge \((t, "Advisee",h)\) should also exist.
Are TransE expressive enough to capture these patterns?
TransR
TransE models translation of any relation in the same embedding space.
TransR: model entities as vectors in the entity space \(\mathbb{R}^d\) and model each relation as vector in relation space \(\mathbf{r} \in \mathbb{R}^{k}\) with \(\mathbf{M}_r \in \mathbb{R}^{k \times d}\) as the projection matrix.
\(h_{k} = M_r h, t_k=M_r t\)
scoring func: $$ f_r(h,t) = -|| h_k + r - t_k || $$
- DistMult
Entities and relations are vectros in \(\mathbb{R}^k\)
Score func:
Intuition of the score function: Can be viewed as a cosine similarity between \(\mathbf{h} \cdot \mathbf{r}\) and \(\mathbf{t}\)
- ComplEx
model entities and relations as complex vectors in \(\mathbb{C}^k\)
Score func:
