Purpose of paper

[janguck]

When you look at the metrics, you compare real meta data with predicted meta data. What do you do with referrals? Or is this the end of the paper or code?

[lgalke]

Hi @janguck, thanks for your interest in our work.

We do the comparison via identifier tokens. Assume some paper cites 20 other papers, then we randomly drop some of these references and then, the task is to predict exactly those ones that have been artificially removed. We use datasets, in which full citation data is available (such as DOI to DOI links). In terms of recommender systems, we reconstruct item sets while having between hundred-thousands or millions of distinct items.

For the input side, we consider using also additional meta-data such as the title of the citing paper, but the output stays the same. The predicted identifiers are unique and can be resolved to a full meta-data record whenever needed.

Does this answer your question?

[janguck]

Thank you for answer. I'm wondering if there is more than one pmcld, title, type, year, month, file, journal column in the .sql document table.