relation matching

[lxylxyoo]

Thanks for opening source !
I have read the code, but I did not find the relation matching module. The current prediction logic seems to be predicting scores for all entities of KG. Will you provide the code for relation matching， because I am not sure how is the shortest path between head entity and candidate answer defined.

[apoorvumang]

Yes we will provide soon, along with the data and instructions.
Thanks!

[Hy-Choi]

@apoorvumang When do you provide data and instructions?

[IISCAditayTripathi]

@Hy-Choi @lxylxyoo we have added the data and instructions to perform experiments on MetaQA dataset. We will add the same for WebQSP in a couple of days. Please check out the code and data. Thanks.

[Hy-Choi]

@IISCAditayTripathi OK, Thanks.

[Kelaxon]

Hi, @apoorvumang thanks for providing these useful codes and instructions.
I am also trying to implement the relation matching part and have a question regarding pruning candidates (a' \in Nh) in section 4.4.1:

Here, what is the meaning of Nh and how can we obtain it? You mentioned that candidates a' could be obtained in section 4.4. However, I see a' are all possible answers in section 4.4 (which is not pruned). Hope to get your feedback. Thanks.

[apoorvumang]

Hi @Kelaxon

That is a typo. It should be same as in section 4.4 ie a' \in E. We were earlier experimenting with restricting the neighbourhood, which was denoted by N_h, but the final version of EmbedKGQA does not have that restriction.

[Kelaxon]

@apoorvumang Thanks for your reply.

Yes, I found the restricting neighborhood part in getNeighbourhood() in folder RoBERTa. It has one hyper-parameter radius. So, what is the empirical range of it? Besides, according to section 4.4, it actually has two parameters needed to determine: radius and r(gamma in the paper), right?

[RainbowCatSzy]

@apoorvumang When do you provide relation matching code?It's nine months since you said public code last time.I'm very curious about the implementation of relation matching

[apoorvumang]

@RainbowRr Apologies, I have been busy with some other work and sidelining this for way too long. Please keep posted on this thread I will add the implementation in some form or another within the next 3 days.

[RainbowCatSzy]

@RainbowRr Apologies, I have been busy with some other work and sidelining this for way too long. Please keep posted on this thread I will add the implementation in some form or another within the next 3 days.

Thank you very much! Wish you success in your work and research!

[apoorvumang]

@RainbowRr I have added the code. Please let me know if you face any issues

[apoorvumang]

You will need to re-download the data and pretrained_models

[apoorvumang]

Closing for now, please reopen a new issue if you have problems with the implementation