search

koustuvsinha / hier-class

Hierarchical Neural Attention-based Classification
25 stars 8 forks source link

dbp dataset #4

Closed lalalei21 closed 4 years ago

lalalei21 commented 4 years ago

Hi, I download the dbp dataset and find it has only 342,782 documents, which is less than your given in the paper. Is that all?

koustuvsinha commented 4 years ago

Hi @lalalei21 , yes that is the cleaned version of the dataset after cleaning and balancing the classes. I think we may have missed to update the final numbers in the paper - I will update it in our next arxiv revision. Thanks for pointing it out!

lalalei21 commented 4 years ago

@koustuvsinha Thanks for your reply! I have another question. Your code prints "Model Parameters: 50682922", but the experimental results show the 34M parameters. Whether the word embedding parameters are excluded or there are other reasons.