GentleZhu/HEER

Are there pre-trained HEER embeddings for YAGO?

Opened this issue · 4 comments

oapandit commented

Hi,

Your algorithm to learn embeddings is quite appealing for networks like YAGO.
I am particularly interested in using YAGO embeddings based on HEER.
Have you made those embeddings available ?

I can find LINE embeddings which you have provided but not HEER. Please let me know if there are any ? I will highly appreciate.

thanks,
Onkar

GentleZhu commented

Hey Onka,

Can you run the HEER algorithm on YAGO following the instructions? Since people have different versions of YAGO, and we report performance on a processed subset on our side. So I think you can run ours YAGO first and extend it to your use case.

Best,
Qi

oapandit commented

Hi Qi,

Thanks for the reply.
True that, I can run your system but there are some memory constraints on my machine. And size of complete YAGO is 168 G!! So thought of checking with you, if you have stored your embeddings somewhere.

I want to use these embeddings for some downstream NLP task and not for any performance comparison.

On the side note, can I ask you a question ? This is about the LINE embeddings, you have provided. From where can I get the string/word mapping of the key in the embeddings ? For example, I read file with your code to create emb_dict, which has key and 128 dim vector associated to it. Key looks something WO:. I understood that these are the entries from YAGO but where to find actual string (in this particular case organization name) mapping to the identifier.

Or the question with the YAGO.zip which you have provided. These files have entries like
PE:171784 AS:35141 1 , what is PE:171784 ?? where is the mapping for this?
I am really confused between the YAGO preview file entries and these entries. The preview files on the official YAGO website is like, <Jean-Baptiste-Joseph_Gobel> rdf:type <wikicat_People_executed_by_guillotine> . Why is the data different in these two files ?
This might be really stupid question but please help me out!

I highly appreciate your help!
Thanks,
Onkar

GentleZhu commented

I'm sorting out the id mapping and I will upload it soon.

oapandit commented

Thanks for the help!