Disease synonyms/hyponyms via synonym list

[michelole]

Take a look at mesh2solrsyn

[michelole]

I got a valid synonym file with the 2015 MeSH version:

go run mesh2solrsyn.go d2015.bin < mtrees2015.bin > synonyms.txt

[michelole]

@steschu63 has provided a python script to create a Solr synonym file out of MeSH 2017. Use it.

[michelole]

This relates to #85 .

[michelole]

The synonym list worsened results from 0,7693 to 0,4347.

It could still benefit, however, from most-fields (#97).

[steschu63]

Hi Michel, ​

The synonym list worsened results from 0,7693 to 0,4347. ​ ​

​Do you mean the MeSH synonym and hyponym list? ​ ​I suppose, with the list we found just additional PMIDs that were not (yet) part of the gold standard. The question is now to find out whether these new PMIDs are relevant. - Stefan

[steschu63]

I would restrict the use of the synonym list to diseases and comorbidities. - Stefan 2017-07-29 19:03 GMT+02:00 Michel Oleynik <[email protected]>:

…

The synonym list worsened results from 0,7693 to 0,4347. It could still benefit, however, from most-fields (#97 <#97>). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#107 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMTHtXD0j-vJ-1OYXQAP9xNUJzIWuiDnks5sS2XugaJpZM4OaTMm> .

-- *STEFAN SCHULZ <http://purl.org/steschu> * (Univ.-Prof. Dr. med.) *Institute for Medical Informatics, Statistics and Documentation <http://www.medunigraz.at/imi/en/index.php>Medical University of Graz <https://www.medunigraz.at/en/>* Auenbruggerplatz 2/V, 8036 Graz (Austria) +43 316 38516939 / +43 699 15096270 / +43 316 38513201 / Skype: stschulz Home: Wiener Str. 5/4, 8020 Graz (Austria)

[michelole]

Hi @steschu63 !

We're using it only for diseases (comorbidities were never explored).

I took a manual look at the results from the worse topics (see gold standard where source column is set to "Synonym list") and the results are still not promising. It's basically matching first rare diseases or papers about virus (?) that rank higher because of tf-idf.

The metrics refer already to the gold standard with these samples.

Ps.: All your comments are public in Github when you reply by email. ;)