HermannKroll / NarrativeIntelligence

GNU General Public License v3.0
3 stars 0 forks source link

Drug Overviews: Aggregation by entity classes instead entity ids #152

Open HermannKroll opened 1 year ago

HermannKroll commented 1 year ago

Drug Overview. Search for Metformin. The service will show around 4k Indications with Diabetes Mellitus. But if a user clicks on the button, the search in the Narrative Service will return around 6k publications.

This is due to the following reason: The Narrative Service expands the search by all subclasses of the entity "Diabetes Mellitus" (e.g. type 1 and type 2 diabetes). Drug Overviews don't do such an expansion. Maybe the expansion should be done in the backend to minimize confusion.

HermannKroll commented 1 year ago

The same "issue" occurs in the substitution-based visualization in the Narrative Service http://www.narrative.pubpharm.de/?query=Metformin+treats+%3FX%28Disease%29&data_source=PubMed&visualization=outer_ranking_substitution&sort_frequency_desc=True&sort_year_desc=True&start_pos=0

HermannKroll commented 1 year ago

The question is how many superclasses should be considered. If we consider all it may be far too general, e.g.,

image
HermannKroll commented 12 months ago

Feedback user: Im Narrative Modus zeigt ja die Verbindungslinie eine Zahl der Co-occurences. Intuitiv würde ich das als Anzahl Dokumente verstehen, in denen beide Begriffe zusammen auftauchen. Zum Teil sind dann aber wenn man drauf klickt, die Zahlen viel höher als im Graphen. Beispiel: Long Covid Overview: Long Covid - Infections: Linie 379, Draufklicken 1026 Dokumente. Ist das ein Bug, oder haben die Zahlen eine andere Bedeutung? Oder gibt es einen Indexing lag zwischen den für den Graphen hinterlegten Zahlen und der restlichen Datenbank?

HermannKroll commented 8 months ago

If we would aggregate by class ids, the lists will be flooded by superclasses of diseases. This is not very intuitive for our users. At the moment, we don't have a solution. Stalled.