Question

[b1ackwater3]

Dear XiouGe: It's my honor to read your article "Compounding Geometric Operations for Knowledge Graph Completion", and I would appreciate it if you could reply to me that "Are all three parameters($S,R,T,\hat{S},\hat{R},\hat{T}$) used in the head entity and tail entity different in the image?"

Based on what I see in the paper, there's this passage 「We only show a single plot for the rotation component since the rotation parameter is shared between the head and the tail」. Does this imply that the parameters used for rotation are the same($R = \hat{R}$)?

Also, if you could release the source code for FB15k-237 and WN18RR publicly, I would greatly appreciate it.

[b1ackwater3]

Additional, I would like to ask what is the negative sampling strategy during testing when using the FB15k-237 and WN18RR datasets? In the testing phase when using the OGB dataset, it seems that negative samples are selected from pre-generated samples. Thanks for answering!

[TTYee]

Dear XiouGe: It's my honor to read your article "Compounding Geometric Operations for Knowledge Graph Completion", and I would appreciate it if you could reply to me that "Are all three parameters( S , R , T , S ^ , R ^ , T ^ ) used in the head entity and tail entity different in the image?"

Based on what I see in the paper, there's this passage 「We only show a single plot for the rotation component since the rotation parameter is shared between the head and the tail」. Does this imply that the parameters used for rotation are the same( R = R ^ )?

Also, if you could release the source code for FB15k-237 and WN18RR publicly, I would greatly appreciate it.

Hello, did you successfully reproduce the results on the WN18RR & FB15k-237 dataset?