def sampleEntityQueries(featFile, clusterFile):
queries = {}
for line in open(featFile, 'r'):
split = line.split('\t')
entityDict = ast.literal_eval(split[5])
qsplit = split[0].split()
if len(qsplit) > 2:
queries[split[0].strip()] = entityDict
outFile = open('EntityQueries.sample', 'w')
globalEnt = {}
entTerms = {}
for cluster in loadClusters(clusterFile):
if len(cluster) > 4:
cent = {}
for query in cluster:
if query in queries:
for entry, value in queries[query].items():
if entry not in cent:
cent[entry] = 0.0
cent[entry] += value
entSort = sorted(cent.items(), reverse=True, key=lambda x: x[1])
toPrint = {}
if len(entSort) > 0:
bestEnt = entSort[0][0]
if bestEnt not in globalEnt:
globalEnt[bestEnt] = []
entTerms[bestEnt] = set([])
for query in cluster:
if query in queries and bestEnt in queries[query]:
qsplit = query.split() #getNGramsAsList(query,2) #
for entry in qsplit:
if len(entry) > 2 and entry not in stopSet \
and entry not in bestEnt and entry not in ashleelString:
if entry not in toPrint:
toPrint[entry] = 0.0
toPrint[entry] += 1.0
sortP = sorted(toPrint.items(), reverse=True, key=lambda x: x[1])
fset = set([])
if len(sortP) > 3:
for entry in sortP:
if entry[1] > 1:
fset.add(entry[0])
if len(fset) > 7:
break
covered = 0
for entry in fset:
if entry not in entTerms[bestEnt]:
entTerms[bestEnt].add(entry)
else:
covered += 1.0
if len(fset) > 3 and covered / len(fset) < .20:
globalEnt[bestEnt].append(fset)
#else:
# print covered, len(fset)
for bestEnt, sList in globalEnt.items():
if len(sList) > 1:
for sortP in sList:
outFile.write(bestEnt.encode('utf-8') + '\t' + '\t'.join(sortP) + '\n')
outFile.close()
def main(argv):
#Scorer
coSessOccur = CoOccurrence()
coSessOcMan = CoOcManager(argv[2], coSessOccur, ' ')
tScorer = CoOccurSimScore(coSessOcMan)
cScorer = ScoreClusterTerms()
#vocab = set()
i = 0
prec = {}
mrr = {}
lim = 55
queryList = loadFileInList(argv[5])
termList, termDict = getTermList(queryList)
print len(termList)
added = 0
oracle_prec = 0.0
oracle_mrr = 0.0
for tid, session, viewDocs, clickDocs, cTitle, cSummary in getSessionWithXML(
argv[1]):
query = session[0].strip()
aTerms, rTerms = addedAndRemovedTerms(query, session[1:], termDict)
if len(aTerms) > 0:
prec1, mrr1 = getPrecRecall(termList, aTerms)
added += 1.0
oracle_prec += prec1
oracle_mrr += mrr1
print 'Oracle prec and recall ', oracle_prec / added, oracle_mrr / added
porter = stem.porter.PorterStemmer()
ttype = argv[6]
print ttype
for iFile in os.listdir(argv[3]):
qclusters = loadClusters(argv[3] + '/' + iFile)
clusters, clusIndex = toTerms(qclusters)
print iFile, len(clusters)
prec[iFile] = {}
mrr[iFile] = {}
added = 0.0
i = 1
for tid, session, viewDocs, clickDocs, cTitle, cSummary in getSessionWithXML(
argv[1]):
i += 1
query = session[0].strip()
qSet = getQueryTermsStemmed(query, porter)
print 'Query ', query, qSet
if ttype == 'query':
aTerms, rTerms = addedAndRemovedTerms(query, session[1:], termDict)
elif ttype == 'title':
aTerms = getTerms(cTitle, qSet, termDict, porter, range(
1, len(session) - 1))
else:
aTerms = getTerms(cTitle, qSet, termDict, porter, range(
1, len(session) - 1))
bTerms = getTerms(cSummary, qSet, termDict, porter, range(
1, len(session) - 1))
aTerms = aTerms | bTerms
#aTerms,rTerms = addedAndRemovedTerms(query, session[1:], None )
if len(aTerms) > 0:
terms = cScorer.scoreWithIndex(qSet, clusters, clusIndex, tScorer, lim)
#terms = cScorer.scoreWithClustPos(qSet, clusters,tScorer, lim)
print 'TERMS', '\t', i, '\t', ttype, '\t', iFile, '\t', len(
terms), terms
#for topk in range(1,lim,5):
prec1, mrr1 = getClustPrecMrr(terms, aTerms) # returns a list
print 'METRIC', iFile, i, prec1, mrr1
#print topk , prec1, mrr1
for topk in prec1.keys():
if topk not in prec[iFile]:
prec[iFile][topk] = []
mrr[iFile][topk] = []
prec[iFile][topk].append(prec1[topk])
mrr[iFile][topk].append(mrr1[topk])
#prec[iFile][topk] += prec1
#mrr[iFile][topk] += mrr1
added += 1.0
#if i == 3:
# break
for fName, scoreDict in prec.items():
for pos in scoreDict.keys():
print 'Prec all', fName, pos, len(scoreDict[pos])
total = sum(scoreDict[pos])
prec[fName][pos] = total / added #len(scoreDict[pos])
print 'Prec', fName, pos, prec[fName][pos], total
for fName, scoreDict in mrr.items():
for pos in scoreDict.keys():
print 'Mrr all', fName, pos, len(scoreDict[pos])
total = sum(mrr[fName][pos])
mrr[fName][pos] = total / added #len(scoreDict[pos])
print 'MRR', fName, pos, mrr[fName][pos], total
#for entry in prec.keys():
#for t in prec[entry].keys():
#print 'Prec',entry, t, prec[entry][t], prec[entry][t]/added
#prec[entry][t]/=added
#for entry in mrr.keys():
#for t in mrr[entry].keys():
#print 'Mrr',entry, t, mrr[entry][t], mrr[entry][t]/added
#mrr[entry][t]/=added
print 'Plotting Precision and MRR'
plotMultipleSys(prec, 'No of Terms', 'Prec', argv[4] + 'prec.png',
'Term Prediction Prec Plot')
plotMultipleSys(mrr, 'No of Terms', 'MRR', argv[4] + 'mrr.png',
'Term Prediction MRR Plot')
