jinheonbaek / gen Goto Github PK

Thanks for your excellent work on OOG link prediction. Could you please tell me how to get pre-trained embeddings for my own dataset with entities and relations?

Hi there,
Firstly, thank you for your work.
I have read your paper and I have a few questions to clarify.
Mainly, how do you "... formulate a set of tasks such that the model
learns to generalize over unseen entities, which are simulated using seen entities." Your paper also mentioned sampling a task from the distribution p(T) but how is p(T) obtained? Is it predefined?

In other words, in the aspect of code, how did you pre-process your data such that it is split into meta-train/meta-valid/meta-test triplets?

self.filtered_triplets, self.meta_train_task_triplets, self.meta_valid_task_triplets, self.meta_test_task_triplets, \
        self.meta_train_task_entity_to_triplets, self.meta_valid_task_entity_to_triplets, self.meta_test_task_entity_to_triplets \
            = utils.load_processed_data('./Dataset/processed_data/{}'.format(args.data))

Other than that, do you mind elaborating on "our meta-learning framework can simulate the unseen entities during meta-training." cause I am still a confused by how your model works.

Thanks!!!

Hi,

thanks for your great work first!

I have a question about the pre-trained embedding generated with Distmult. Do we need to ignore "unseen entities" when we pre-train, i.e., remove all the triples which contain unseen entities? Or we just put the whole KG into the pre-training process, and then mask them in inductive/transductive training?

Thanks!

Hi,

I am wondering how you train these methods in your task. As I understand, these three methods use entity pair matching and they do not use embeddings of sparse relations during training and evaluation. Do you also neglect the triple relations (relations in the triples containing unseen entities) while training them?

And also I think Gmatching, MetaR and FSRL are originally using meta-learning framework. What is the difference between, e.g., Gmatching and Gmatching*?

Hi,

Your work is very interesting. Can you explain to me the inductive and transductive setup of data while training and testing? How they are different in your setup in terms of dataset preparation for training and testing.

Thank you for your excellent work!

Unseen entities are split into (meta-)train/val/test. In your code, i found that the triplets in self.meta_train_task_entity_to_triplets, self.meta_val_task_entity_to_triplets and self.meta_test_task_entity_to_triplets are disjoint respectively. Is this just a coincidence or a must? Because i have a question, if entity s is divided into train and entity o into test, the triplet (s,r,o) belongs to which? In this case, the triplets in train and test will not be disjoint.