Open Jean-Baptiste-Camps opened 6 years ago
Another potential issue with this calculation is when the only conflict is of a single VL (with alternative readings) against himself. In which case, the index will be -2…
C’est vrai... Faut qu’on change. Des idées, je te t’en parle dimanche.
Florian Cafiero
Le 11 mai 2018 à 19:03, Jean-Baptiste-Camps notifications@github.com a écrit :
Another potential issue with this calculation is when the only conflict is of a single VL (with alternative readings) against himself. In which case, the index will be -2…
— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub, or mute the thread.
Ok, while adding more tests I stumbled on the same error as in may of last year. A centrality index of -2 for the (rare and very theoric) case where there is a single conflict of a VL with itself. Should we change computation for this ? Or renounce altogether the use of 'alternateReadings' that are causing inconsistencies everywhere, and are very hard to manage ?
@floriancafiero , any thoughts ?
Okay, so, basically, the most easy solution I think of is switching from
deg(u) / e - deg(u)
to
\frac{deg(u)}{\sum_{v \in V} deg(v)}
which would normalise on [0;1].
Another solution could be to look on the various centrality measures, and test to see if some could be interesting for us as well.
For now, we use the index offered in the paper
I'm asking myself questions on two aspects:
e = deg(u)
? For now, the code on this point is a bit of a hack. If the result is infinite, I normalise it to 2… We could always do,deg(u) / e
(perhaps better thandeg(u)/ e - deg(u) + 1
), which would normalise the result on 0 … 1 ?The current code, that can be really enhanced: