def __init__(self, queryFile, candidatePath, mu, corpusFile, sigma, lamda):
self.query = {}
self.candidate = candidatePath
self.tweet = {}
self.mu = mu
self.sigma = sigma #similarity threshold
self.lamda = lamda #cluster threshold
self.jaccInstance = Jaccard()
self.klInstance = Distance(mu, corpusFile)
print "corpus read done!"
def __init__(self, queryFile, candidatePath, mu, corpusFile, sigma, lamda):
self.query = {}
self.candidate = candidatePath
self.tweet = {}
self.mu = mu
self.sigma = sigma #similarity threshold
self.lamda = lamda #cluster threshold
self.jaccInstance = Jaccard()
self.klInstance = Distance(mu, corpusFile)
print "corpus read done!"
class Cluster:
def __init__(self, queryFile, candidatePath, mu, corpusFile, sigma, lamda):
self.query = {}
self.candidate = candidatePath
self.tweet = {}
self.mu = mu
self.sigma = sigma #similarity threshold
self.lamda = lamda #cluster threshold
self.jaccInstance = Jaccard()
self.klInstance = Distance(mu, corpusFile)
print "corpus read done!"
def write(self, writePath, alpha, yibuson):
writeFile = writePath + "res.rmSW4.59S" + str(self.sigma) + ".L" + str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson)
result = open(writeFile, "w+")
log = open(writePath + "log.rmSW4.59S" + str(self.sigma) + ".L" + str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
log.write("Qid\tclusterCount\ttweetCount\n")
log1 = open(writePath + "qidWidKL.rmSW4.59S" + str(self.sigma) + ".L" + str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
log2 = open(writePath + "widWidKL.rmSW4.59S" + str(self.sigma) + ".L" + str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
log3 = open(writePath + "jaccScore.rmSW4.59S" + str(self.sigma) + ".L" + str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
num = 1
files = []
while(num <= 55):
files.append(str(num) + ".res.content.all")
num += 1
for file in files:
#remember to make them initial on each query
self.curQid = -1
self.cluster = []
self.qidWidKL = {}
self.qidWidMax = 0
self.qidWidMin = 999
self.widWidKL = {}
self.widWidMax = 0
self.widWidMin = 999
self.widScore = {}
self.jacc = {}
self.jaccMax = 0
self.jaccMin = 1
self.resultList = []
readPath = self.candidate + file
with open(readPath, "r") as fin:
for i, line in enumerate(fin):
qid, Qid, wid, rank, score, runName, wcontent, qcontent = line.strip().split("\t")
self.query[qid] = qcontent
#first time selection
if(float(score) < 4.59):
if not self.cluster:
print "break out of 4.59, ", file, " , empty cluster!"
exit()
break
self.tweet[wid] = wcontent
self.curQid = qid
#calculate qidWidKL
similarity = self.klInstance.kl(self.query[qid], wcontent)
#calculate jaccard score
jaccScore = self.jaccInstance.jaccardScore(qcontent, wcontent)
#if similarity <= self.sigma and jaccScore >= yibuson:
if similarity <= self.sigma:
#set self.qidWidKL
self.qidWidKL[qid+"-"+wid] = similarity
if (similarity > self.qidWidMax):
self.qidWidMax = similarity
if (similarity < self.qidWidMin):
self.qidWidMin = similarity
#set self.jacc
self.jacc[qid+"-"+wid] = jaccScore
if self.jaccMax < jaccScore:
self.jaccMax = jaccScore
if self.jaccMin > jaccScore:
self.jaccMin = jaccScore
#calculate widWidKL
if not self.cluster:
self.cluster.append([wid])
self.widWidKL[wid] = []
else:
self.__clustering(wid)
if (i % 100) == 0:
print file, " => ", i
log1.write(str(self.qidWidMin) + "\t" + str(self.qidWidMax) + "\n")
for key in self.qidWidKL:
log1.write(key + "\t" + str(self.qidWidKL[key]) + "\n")
log2.write(str(self.widWidMin) + "\t" + str(self.widWidMax) + "\n")
for key in self.widWidKL:
for i in range(len(self.widWidKL[key])):
for widKey in self.widWidKL[key][i]:
log2.write(key + "-" + widKey + "\t" + str(self.widWidKL[key][i][widKey]) + "\n")
log3.write(str(self.jaccMin) + "\t" + str(self.jaccMax) + "\n")
for key in self.jacc:
log3.write(key + "\t" + str(self.jacc[key]) + "\n")
rankInstance = Rank(self.widWidKL, self.widWidMin, self.widWidMax)
#self.widScore = rankInstance.textRank()
self.widScore = rankInstance.combinedCov(alpha, self.cluster, self.tweet)
#log info
clusterCount = len(self.cluster)
tweetCount = 0
#select one wid from each cluster
for i in range(len(self.cluster)):
maxScore = 0
bestWid = -1
tweetCount += len(self.cluster[i])
for wid in self.cluster[i]:
if self.widScore[wid] > maxScore:
#select min query-tweet kl score
#key = str(self.curQid) + "-" + wid
#if self.qidWidKL[key] > maxScore:
#maxScore = self.qidWidKL[key]
maxScore = self.widScore[wid]
bestWid = wid
self.resultList.append(bestWid)
#write log info
log.write("MB" + self.curQid + "\t" + str(clusterCount) + "\t" + str(tweetCount) + "\n")
#write result
for wid in self.resultList:
result.write("MB" + self.curQid + "\t" + "Q0\t" + wid + "\t1\t1\tYAO\n")
def __clustering(self, wid):
minScore = 999
index = -1
wcontent = self.tweet[wid]
for i in range(len(self.cluster)):
for cwid in self.cluster[i]:
ccontent = self.tweet[cwid]
score = self.klInstance.kl(self.tweet[wid], self.tweet[cwid])
#print i
if wid in self.widWidKL:
self.widWidKL[wid].append({cwid: score})
else:
self.widWidKL[wid] = [{cwid: score}]
if cwid in self.widWidKL:
self.widWidKL[cwid].append({wid: score})
else:
self.widWidKL[cwid] = [{wid: score}]
#select miniScore, that is the most similar value
if score < minScore:
minScore = score
index = i
#record self.widWidMax & self.widWidMin
if score < self.widWidMin:
self.widWidMin = score
if score > self.widWidMax:
self.widWidMax = score
#put wid into the cluster
#a new cluster
if minScore > self.lamda:
self.cluster.append([wid])
#print self.cluster
#add to a highest similarity cluster
else:
self.cluster[index].append(wid)
class Cluster:
def __init__(self, queryFile, candidatePath, mu, corpusFile, sigma, lamda):
self.query = {}
self.candidate = candidatePath
self.tweet = {}
self.mu = mu
self.sigma = sigma #similarity threshold
self.lamda = lamda #cluster threshold
self.jaccInstance = Jaccard()
self.klInstance = Distance(mu, corpusFile)
print "corpus read done!"
def write(self, writePath, alpha, yibuson):
writeFile = writePath + "res.rmSW4.59S" + str(self.sigma) + ".L" + str(
self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson)
result = open(writeFile, "w+")
log = open(
writePath + "log.rmSW4.59S" + str(self.sigma) + ".L" +
str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
log.write("Qid\tclusterCount\ttweetCount\n")
log1 = open(
writePath + "qidWidKL.rmSW4.59S" + str(self.sigma) + ".L" +
str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
log2 = open(
writePath + "widWidKL.rmSW4.59S" + str(self.sigma) + ".L" +
str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
log3 = open(
writePath + "jaccScore.rmSW4.59S" + str(self.sigma) + ".L" +
str(self.lamda) + ".A" + str(alpha) + ".NJ" + str(yibuson), "w+")
num = 1
files = []
while (num <= 55):
files.append(str(num) + ".res.content.all")
num += 1
for file in files:
#remember to make them initial on each query
self.curQid = -1
self.cluster = []
self.qidWidKL = {}
self.qidWidMax = 0
self.qidWidMin = 999
self.widWidKL = {}
self.widWidMax = 0
self.widWidMin = 999
self.widScore = {}
self.jacc = {}
self.jaccMax = 0
self.jaccMin = 1
self.resultList = []
readPath = self.candidate + file
with open(readPath, "r") as fin:
for i, line in enumerate(fin):
qid, Qid, wid, rank, score, runName, wcontent, qcontent = line.strip(
).split("\t")
self.query[qid] = qcontent
#first time selection
if (float(score) < 4.59):
if not self.cluster:
print "break out of 4.59, ", file, " , empty cluster!"
exit()
break
self.tweet[wid] = wcontent
self.curQid = qid
#calculate qidWidKL
similarity = self.klInstance.kl(self.query[qid], wcontent)
#calculate jaccard score
jaccScore = self.jaccInstance.jaccardScore(
qcontent, wcontent)
#if similarity <= self.sigma and jaccScore >= yibuson:
if similarity <= self.sigma:
#set self.qidWidKL
self.qidWidKL[qid + "-" + wid] = similarity
if (similarity > self.qidWidMax):
self.qidWidMax = similarity
if (similarity < self.qidWidMin):
self.qidWidMin = similarity
#set self.jacc
self.jacc[qid + "-" + wid] = jaccScore
if self.jaccMax < jaccScore:
self.jaccMax = jaccScore
if self.jaccMin > jaccScore:
self.jaccMin = jaccScore
#calculate widWidKL
if not self.cluster:
self.cluster.append([wid])
self.widWidKL[wid] = []
else:
self.__clustering(wid)
if (i % 100) == 0:
print file, " => ", i
log1.write(str(self.qidWidMin) + "\t" + str(self.qidWidMax) + "\n")
for key in self.qidWidKL:
log1.write(key + "\t" + str(self.qidWidKL[key]) + "\n")
log2.write(str(self.widWidMin) + "\t" + str(self.widWidMax) + "\n")
for key in self.widWidKL:
for i in range(len(self.widWidKL[key])):
for widKey in self.widWidKL[key][i]:
log2.write(key + "-" + widKey + "\t" +
str(self.widWidKL[key][i][widKey]) + "\n")
log3.write(str(self.jaccMin) + "\t" + str(self.jaccMax) + "\n")
for key in self.jacc:
log3.write(key + "\t" + str(self.jacc[key]) + "\n")
rankInstance = Rank(self.widWidKL, self.widWidMin, self.widWidMax)
#self.widScore = rankInstance.textRank()
self.widScore = rankInstance.combinedCov(alpha, self.cluster,
self.tweet)
#log info
clusterCount = len(self.cluster)
tweetCount = 0
#select one wid from each cluster
for i in range(len(self.cluster)):
maxScore = 0
bestWid = -1
tweetCount += len(self.cluster[i])
for wid in self.cluster[i]:
if self.widScore[wid] > maxScore:
#select min query-tweet kl score
#key = str(self.curQid) + "-" + wid
#if self.qidWidKL[key] > maxScore:
#maxScore = self.qidWidKL[key]
maxScore = self.widScore[wid]
bestWid = wid
self.resultList.append(bestWid)
#write log info
log.write("MB" + self.curQid + "\t" + str(clusterCount) + "\t" +
str(tweetCount) + "\n")
#write result
for wid in self.resultList:
result.write("MB" + self.curQid + "\t" + "Q0\t" + wid +
"\t1\t1\tYAO\n")
def __clustering(self, wid):
minScore = 999
index = -1
wcontent = self.tweet[wid]
for i in range(len(self.cluster)):
for cwid in self.cluster[i]:
ccontent = self.tweet[cwid]
score = self.klInstance.kl(self.tweet[wid], self.tweet[cwid])
#print i
if wid in self.widWidKL:
self.widWidKL[wid].append({cwid: score})
else:
self.widWidKL[wid] = [{cwid: score}]
if cwid in self.widWidKL:
self.widWidKL[cwid].append({wid: score})
else:
self.widWidKL[cwid] = [{wid: score}]
#select miniScore, that is the most similar value
if score < minScore:
minScore = score
index = i
#record self.widWidMax & self.widWidMin
if score < self.widWidMin:
self.widWidMin = score
if score > self.widWidMax:
self.widWidMax = score
#put wid into the cluster
#a new cluster
if minScore > self.lamda:
self.cluster.append([wid])
#print self.cluster
#add to a highest similarity cluster
else:
self.cluster[index].append(wid)
