Support for users of CCG software closed 7-27-20

[rupen AT cs.vt.edu]

Hi,

I am using wikifier 2.0 on location/place-names strings in my data. For a
particular case described below, I would like some help:

input string: "Distrito Federal, Mexico"

output html: '<a class="wiki"
href="http://en.wikipedia.org/wiki/Distrito_Federal"; cat="">Distrito Federal</
a> , Mexico'

Here, hyperlinked wiki page is a disambiguation page. And further, among the
wikified entities, for "Distrito Federal" disambiguation candidates in output
contains both:
 a . {u'RankerScore': u'0.3349157584747268',
         u'WikiTitle': u'Mexican_Federal_District',
         u'WikiTitleID': u'222236'},
b.   {u'RankerScore': u'0.1631062757211779',
         u'WikiTitle': u'Distrito_Federal',
         u'WikiTitleID': u'401300'}

Now, even though score of "Mexican_Federal_District" is higher than
"Distrito_Federal", the latter was choosen.

Question: Where/How in the code this choice was made. Perhaps if I can
identify how the logic works I can modify it too suit my research needs.
Alternatively, if there is particular setting in the config that I can use to
correct this.

thanks in advance.

ps:
1. I am using BEAM_SEARCH as ILPSolver. I wasn't able to get Gurobi 5.5 work,
primarliy that Gurobi guys have discontinued academic license validation for
v5.5 that forces you to use their latest version 6.0, further that has its own
problems since maven build doesnt support the latest Gurobi version.
2. My Config settings:

minLinkability = -1.0
minSurfaceLen = 0.0
EXCLUDE_DISAMBIGUATION_PAGE = false
GENERATE_CONTEXTUAL_CANDIDATES = true
USE_LEXICAL_SEARCH_HEURSTICS = true
USE_COREF = true
DISABLE_LOC_COREF = false
USE_RELATIONAL_INFERENCE = true
RESOLVE_NOMINAL_COREF = true
useLexicalFeaturesNaive = false
disallowEntityOverlap = false
useNestedNER = true