def parse_variable_metadata(varfile):
dic = load_acronym('acronym_list.txt')
with open(varfile, 'rb') as f:
varcontent = f.read().splitlines()
varnamelist = []
varfieldlist = []
orgvarlist = []
for i in range(len(varcontent)):
#print varcontent[i]
orgvarlist.append(varcontent[i])
cstr = varcontent[i].split('|||')
#process varname info
namestr = cstr[0].split('/')
cname = []
for j in range(len(namestr)):
#cstr = convert_CamelCase2WhiteSpace(namestr[j])
namestr[j] = nlp_phrasecleaning(namestr[j], dic)
cname.append(nlp_StopStemLemm(namestr[j]))
varnamelist.append(cname)
#process description info
cstr[1] = nlp_phrasecleaning(cstr[1], dic)
varfieldlist.append(nlp_StopStemLemm(cstr[1]))
return varnamelist, varfieldlist, orgvarlist
def parse_GCMD_Keywords(keywordfile):
dic = load_acronym('acronym_list.txt')
#assuming keywords for a data set from a file such as xxx_keyword_original.txt
with open(keywordfile, 'rb') as f:
keylist = f.read().splitlines()
keylist = list(set(keylist))
org_keywordlist = []
#get a set of keywords for this dataset
fullkeywordlist = []
partialkeywordlist = []
for i in range(len(keylist)):
ckeylist = keylist[i]
ckeylist = ckeylist.rstrip()
ckeylist = ckeylist.lstrip()
org_keywordlist.append(ckeylist)
strlist = ckeylist.split('->')
for j in range(len(strlist)):
slist = strlist[j].split('/')
for k in range(len(slist)):
fullkeywordlist.append(slist[k])
plist = slist[k].split()
for m in range(len(plist)):
partialkeywordlist.append(plist[m])
#nlp processing (removing stop word, stem and lemmatize)
fullkeywordlist = list(set(fullkeywordlist))
partialkeywordlist = list(set(partialkeywordlist))
fullwordlist = []
for i in range(len(fullkeywordlist)):
cstr = nlp_phrasecleaning(fullkeywordlist[i], dic)
fullwordlist.append(nlp_StopStemLemm(cstr))
partialwordlist = []
for i in range(len(partialkeywordlist)):
cstr = nlp_phrasecleaning(partialkeywordlist[i], dic)
partialwordlist.append(nlp_StopStemLemm(cstr))
#cleaning original keywordlist for duplicate
org_keywordlist = list(set(org_keywordlist))
print "FF: ", fullwordlist
print "JJ: ", partialwordlist
print "KK: ", org_keywordlist
return list(set(fullwordlist)), list(set(partialwordlist)), org_keywordlist
def parse_GCMD_Keywords_by_list(keylist):
dic = load_acronym('acronym_list.txt')
org_keywordlist = []
#get a set of keywords for this dataset
fullkeywordlist = []
partialkeywordlist = []
for i in range(len(keylist)):
ckeylist = keylist[i]
org_keywordlist.append(ckeylist)
strlist = ckeylist.split('->')
for j in range(len(strlist)):
slist = strlist[j].split('/')
for k in range(len(slist)):
fullkeywordlist.append(slist[k])
plist = slist[k].split()
for m in range(len(plist)):
partialkeywordlist.append(plist[m])
#nlp processing (removing stop word, stem and lemmatize)
fullkeywordlist = list(set(fullkeywordlist))
partialkeywordlist = list(set(partialkeywordlist))
fullwordlist = []
for i in range(len(fullkeywordlist)):
cstr = nlp_phrasecleaning(fullkeywordlist[i], dic)
fullwordlist.append(nlp_StopStemLemm(cstr))
partialwordlist = []
for i in range(len(partialkeywordlist)):
cstr = nlp_phrasecleaning(partialkeywordlist[i], dic)
partialwordlist.append(nlp_StopStemLemm(cstr))
return list(set(fullwordlist)), list(set(partialwordlist)), org_keywordlist
def map_var_2_keyword(keywordlist, varfield, keylist1, keylist2, var_meta, field_meta, \
idf_varlist_k1, idf_fieldlist_k1, idf_varlist_k2, idf_fieldlist_k2):
dic = load_acronym('acronym_list.txt')
num_var = len(var_meta)
num_keywords = len(keywordlist)
ovreall_score = []
#process each keywords
for i in range(num_keywords): #each original science keywords
#print keywordlist[i]
cscorelist = [] #score of each variable matched to this keyword
ckey = keywordlist[i] #current keyword
keylist = ckey.split(
'->') #split current keyword, on various levels of hierarchy
#print keylist
kwd = [
] #keep hierarchy of keywords (as indexed in list), processed by nlp
#as does earlier for variable text for consistency
for j in range(len(keylist)):
cstr = nlp_phrasecleaning(keylist[j], dic)
cstr = nlp_StopStemLemm(cstr)
kwd.append(cstr)
#print kwd
nkwd = float(len(kwd))
#now score each variable based on text matching using tf-idf scheme
var_k1_score = [] #score based on variable name and science keyword k1
for k in range(num_var): #scoring for each variable
cvar = var_meta[k] #variable text for
#print cvar
nlev = float(
len(cvar)
) #number of levels of this variable, the lowest set weight of 1.0
cscore = 0.0 #score
for m in range(
len(cvar)
): #for each level of variable (possible from nc4, hdf5)
cstr = cvar[m] #current level of variable name component
for n in range(len(
kwd)): #try to match to each keyword with hierarchy
tf = cstr.count(
kwd[n]
) #count number of current keyword in the varname
idf = idf_varlist_k1.get(
kwd[n], 0.0) #get idf for kwd[n] word, default 0.0
cscore = cscore + tf * idf * (m + 1) * (n + 1) / (
nkwd * nlev
) #tf-idf is weighted by the hierarchy level of keyword and variable
var_k1_score.append(cscore)
#print var_k1_score
#now score each variable field based on text matching using tf-idf scheme
field_k1_score = [
] #score based on variable name and science keyword k1
for k in range(
num_var
): #scoring for each variable field, same number as variable name
fvar = field_meta[k] #field text for
#print fvar
cscore = 0.0 #score
for n in range(
len(kwd)): #try to match to each keyword with hierarchy
tf = fvar.count(
kwd[n]) #count number of current keyword in the varname
idf = idf_fieldlist_k1.get(
kwd[n], 0.0) #get idf for kwd[n] word, default 0.0
cscore = cscore + tf * idf * (
n + 1
) / nkwd #tf-idf is weighted by the hierarchy level of keyword and variable
field_k1_score.append(cscore)
#print field_k1_score
#now score each variable based on text matching using tf-idf scheme using keyword 2 list
#need to regenerate current keyword list by decomposing keyword phrase into words
elem_kwd = [] #element words (not phrase)
for kk in range(len(kwd)):
cstr = kwd[kk]
carray = cstr.split()
for mm in range(len(carray)):
elem_kwd.append(carray[mm])
elem_kwd = list(set(elem_kwd))
var_k2_score = [] #score based on variable name and science keyword k1
for k in range(num_var): #scoring for each variable
cvar = var_meta[k] #variable text for
nlev = float(
len(cvar)
) #number of levels of this variable, the lowest set weight of 1.0
cscore = 0.0 #score
for m in range(
len(cvar)
): #for each level of variable (possible from nc4, hdf5)
cstr = cvar[m] #current level of variable name component
for n in range(len(elem_kwd)
): #try to match to each keyword with hierarchy
tf = cstr.count(
elem_kwd[n]
) #count number of current keyword in the varname
idf = idf_varlist_k2.get(
elem_kwd[n],
0.0) #get idf for kwd[n] word, default 0.0
cscore = cscore + tf * idf * (
m + 1
) / nlev #tf-idf is weighted by the hierarchy level of variable
var_k2_score.append(cscore)
#print var_k2_score
field_k2_score = [
] #score based on variable name and science keyword k1
for k in range(
num_var
): #scoring for each variable field, same number as variable name
fvar = field_meta[k] #field text for
#print fvar
cscore = 0.0 #score
for n in range(len(
elem_kwd)): #try to match to each keyword with hierarchy
tf = fvar.count(
elem_kwd[n]
) #count number of current keyword in the varname
idf = idf_fieldlist_k2.get(
elem_kwd[n], 0.0) #get idf for kwd[n] word, default 0.0
cscore = cscore + tf * idf #tf-idf
field_k2_score.append(cscore)
#now need to combine scores from the 4 categories by assigning weights(significance)
#set wt1 = 0.5, (subjectively), meaning score from field metadata is half-credited of from variable name
#set wt2 = 0.1, (subjectively), meaning score from phrase (exact match) gives 10 times credit as much as from words
wt1 = 0.5
wt2 = 0.1
for kk in range(num_var):
f_score = var_k1_score[kk] + wt1 * field_k1_score[kk] + \
wt2*var_k2_score[kk] + wt1 * wt2* field_k2_score[kk]
cscorelist.append(f_score)
ovreall_score.append(cscorelist)
#print cscorelist
rank = sorted(range(len(cscorelist)), key=lambda k: cscorelist[k])
rank.reverse()
# for kk in range(num_var):
# idx = rank[kk]
# if cscorelist[idx] > 0.0:
# print ' ' + varfield[idx], cscorelist[idx]
#print out variables that have score > 0.0
return ovreall_score
