[open-science] New version of Open Knowledge Maps

[Peter Kraker]

Hi Roman,

thanks - and good to hear from you! See my answers inline:

On 23/03/2017 17:57, Roman Gurinovich wrote:
> Two questions:
>
> 1. Could you describe a little, how deeply search dives in the text of 
> the paper?
>
>     Options that come to mind:
>     1. Graphs are based on terms from papers metadata, for example
>     'MeSH terms' from Pubmed xml;
>     2. All words from the papers' fulltext are taken into account
>     3. It depends on Pubmed/BASE API results
>

It's option #3. We are using titles, abstracts, authors, keywords and 
publication venue as input for the similarity matrix (you can see for 
yourself in our connector scripts [1] and [2]; everything that's 
returned as part of the list member "text" will be used in the 
similarity matrix). There is a version for PLOS [3], which uses the 
fulltext; but we moved to metadata when we were featured on the 
Hackernews frontpage, as it is much faster to compute on the fly. PubMed 
and BASE do not offer fulltext anyway (with the ecxeption of PMC).

[1] 
https://github.com/OpenKnowledgeMaps/Headstart/blob/master/server/preprocessing/other-scripts/base.R
[2] 
https://github.com/OpenKnowledgeMaps/Headstart/blob/master/server/preprocessing/other-scripts/pubmed.R
[3] 
https://github.com/OpenKnowledgeMaps/Headstart/blob/master/server/preprocessing/other-scripts/rplos.R

> 2. Are you planning API to retrieve graphs in the form of RDF of 
> similar? Like, querying 'mTOR' and receiving all connections between 
> terms in the relevant papers.

Yes, definitely. An API is part of our roadmap: 
https://github.com/OpenKnowledgeMaps/open-discovery/blob/master/roadmap.md#broad-overview

Best,
Peter
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.okfn.org/pipermail/open-science/attachments/20170324/5d3f79ba/attachment-0003.html>