def runLBP(self):
threadedLBP = LBP(self.graph)
threadedLBP.doBeliefPropagationIterative(self.limit)
(fakeUsers, honestUsers, unclassifiedUsers, badProducts,
goodProducts, unclassifiedProducts, fakeReviewEdges,
realReviewEdges, unclassifiedReviewEdges) = threadedLBP.calculateBeliefVals()
self.to_be_removed_usr_bnss_edges = set([(threadedLBP.getEdgeDataForNodes(*edge).getUserId(),\
threadedLBP.getEdgeDataForNodes(*edge).getBusinessID())\
for edge in fakeReviewEdges])
print len(self.to_be_removed_usr_bnss_edges)
def calculateInterestingBusinessStatistics(cross_time_graphs, not_mergeable_businessids, bnss_score_all_time_map):
interesting_bnss_across_time = set([bnss_key for time_key in not_mergeable_businessids for bnss_key in not_mergeable_businessids[time_key]])
bnss_score_across_time_with_interestingMarked = dict()
for bnss_key in interesting_bnss_across_time:
score_across_time_with_intersting_marked = { time_key:(bnss_score_all_time_map[bnss_key][time_key],(0,0),True) \
if time_key in bnss_score_all_time_map[bnss_key] and time_key in not_mergeable_businessids \
and bnss_key in not_mergeable_businessids[time_key]\
else '-' if time_key not in bnss_score_all_time_map[bnss_key] \
else (bnss_score_all_time_map[bnss_key][time_key],(0,0),False) \
for time_key in cross_time_graphs.iterkeys()}
bnss_score_across_time_with_interestingMarked[bnss_key] = score_across_time_with_intersting_marked
for time_key in cross_time_graphs:
graph = cross_time_graphs[time_key]
dummy_lbp = LBP(graph)
(fakeUsers,honestUsers,unclassifiedUsers,\
badProducts,goodProducts,unclassifiedProducts,\
fakeReviewEdges,realReviewEdges,unclassifiedReviewEdges) = dummy_lbp.calculateBeliefVals()
for fakeReviewEdge in fakeReviewEdges:
(siaObject1,siaObject2) = fakeReviewEdge
if siaObject1.getNodeType() == SIAUtil.PRODUCT:
bnssIdFromEdge = siaObject1.getId()
if bnssIdFromEdge in bnss_score_across_time_with_interestingMarked:
score,(edge_sentiment_negative,edge_sentiment_positive),isInteresting = bnss_score_across_time_with_interestingMarked[bnssIdFromEdge][time_key]
if dummy_lbp.getEdgeDataForNodes(*fakeReviewEdge).getReviewSentiment() == SIAUtil.REVIEW_TYPE_NEGATIVE:
edge_sentiment_negative+=1
else:
edge_sentiment_positive+=1
bnss_score_across_time_with_interestingMarked[bnssIdFromEdge][time_key] = score,(edge_sentiment_negative,edge_sentiment_positive),isInteresting
else:
bnssIdFromEdge = siaObject2.getId()
if bnssIdFromEdge in bnss_score_across_time_with_interestingMarked:
score,(edge_sentiment_negative,edge_sentiment_positive),isInteresting = bnss_score_across_time_with_interestingMarked[bnssIdFromEdge][time_key]
if dummy_lbp.getEdgeDataForNodes(*fakeReviewEdge).getReviewSentiment() == SIAUtil.REVIEW_TYPE_NEGATIVE:
edge_sentiment_negative+=1
else:
edge_sentiment_positive+=1
bnss_score_across_time_with_interestingMarked[bnssIdFromEdge][time_key] = score,(edge_sentiment_negative,edge_sentiment_positive),isInteresting
print bnss_score_across_time_with_interestingMarked
def runParentLBPAndCompareStatistics(certifiedFakesFromTemporalAlgo,certifiedRealFromTemporalAlgo, parent_graph):
print "------------------------------------Running Parent LBP along with all Time Edges--------------------------------------"
# run LBP on a non temporal full graph for comparison
parent_lbp = LBP(parent_graph)
parent_lbp.doBeliefPropagationIterative(50)
(parent_lbp_fakeUsers, parent_lbp_honestUsers,parent_lbp_unclassifiedUsers,\
parent_lbp_badProducts, parent_lbp_goodProducts, parent_lbp_unclassifiedProducts,\
parent_lbp_fakeReviewEdges, parent_lbp_realReviewEdges, parent_lbp_unclassifiedReviewEdges) = parent_lbp.calculateBeliefVals()
print "-----------------------------------------------Statistics------------------------------------------------------------------"
fakeReviewsInParentLBP = set([parent_lbp.getEdgeDataForNodes(*edge).getId() for edge in parent_lbp_fakeReviewEdges])
realReviewsInParentLBP = set([parent_lbp.getEdgeDataForNodes(*edge).getId() for edge in parent_lbp_realReviewEdges])
fakeReviewsFromYelp = set([parent_lbp.getEdgeDataForNodes(*edge).getId() for edge in parent_graph.edges()\
if not parent_lbp.getEdgeDataForNodes(*edge).isRecommended()] )
realReviewsFromYelp = set([parent_lbp.getEdgeDataForNodes(*edge).getId() for edge in parent_graph.edges()\
if parent_lbp.getEdgeDataForNodes(*edge).isRecommended()] )
fakeReviewsFromTemporalAlgo = set([parent_lbp.getReviewIdsForUsrBnssId(usrId, bnssId) for (usrId,bnssId) in certifiedFakesFromTemporalAlgo])
realReviewsFromTemporalAlgo = set([parent_lbp.getReviewIdsForUsrBnssId(usrId, bnssId) for (usrId,bnssId) in realFromTemporalAlgo])
totalReviews = len([egde for egde in parent_graph.edges()])
#Accuracy
print 'Fake Reviews Temporal Algo',len(fakeReviewsFromTemporalAlgo)
print 'Fake Reviews LBP',len(fakeReviewsInParentLBP)
print 'Fake Reviews Yelp', len(fakeReviewsFromYelp)
print 'Real Reviews Temporal Algo',len(realReviewsFromTemporalAlgo)
print 'Real Reviews LBP',len(realReviewsInParentLBP)
print 'Real Reviews Yelp', len(realReviewsFromYelp)
print 'Intersection of FakeReviews between Yelp with TemporalLBP:', len(fakeReviewsFromYelp&fakeReviewsFromTemporalAlgo)
print 'Intersection of FakeReviews between Yelp with LBP:', len(fakeReviewsFromYelp&fakeReviewsInParentLBP)
print 'Intersection of FakeReviews between Temporal LBP with LBP:', len(fakeReviewsFromTemporalAlgo&fakeReviewsInParentLBP)
print 'Intersection FakeReviews Across Yelp,Temporal and LBP',len(fakeReviewsFromTemporalAlgo&fakeReviewsInParentLBP&fakeReviewsFromYelp)
print 'Intersection of RealReviews between Yelp with TemporalLBP:', len(realReviewsFromYelp&realReviewsFromTemporalAlgo)
print 'Intersection of RealReviews between Yelp with LBP:', len(realReviewsFromYelp&realReviewsInParentLBP)
print 'Intersection of RealReviews between Temporal LBP with LBP:', len(realReviewsFromTemporalAlgo&realReviewsInParentLBP)
print 'Intersection RealReviews Across Yelp,Temporal and LBP',len(realReviewsFromTemporalAlgo&realReviewsInParentLBP&fakeReviewsFromYelp)
print 'Fake Review - Yelp-TemporalLBP:',len(fakeReviewsFromYelp-fakeReviewsFromTemporalAlgo)
print 'Fake Reviews - TemporalLBP-Yelp:',len(fakeReviewsFromTemporalAlgo-fakeReviewsFromYelp)
print 'Fake Reviews Yelp-LBP:', len(fakeReviewsFromYelp-fakeReviewsInParentLBP)
print 'Fake Reviews LBP-Yelp:', len(fakeReviewsInParentLBP-fakeReviewsFromYelp)
print 'Fake Reviews Temporal LBP-LBP:', len(fakeReviewsFromTemporalAlgo-fakeReviewsInParentLBP)
print 'Fake Reviews LBP-TemporalLBP:', len(fakeReviewsInParentLBP-fakeReviewsFromTemporalAlgo)
trueNegativesTemporalAlgo = len(realReviewsFromYelp&realReviewsFromTemporalAlgo)
truePositivesTemporalAlgo = len(fakeReviewsFromYelp&fakeReviewsFromTemporalAlgo)
trueNegativesLBP = len(realReviewsFromYelp&realReviewsInParentLBP)
truePositivesLBP = len(fakeReviewsFromYelp&fakeReviewsInParentLBP)
accuracyOfTemporalAlgo = (truePositivesTemporalAlgo + trueNegativesTemporalAlgo)/totalReviews
accuracyOfLBP = (truePositivesLBP + trueNegativesLBP)/totalReviews
precisionOfTemporalAlgo = len(fakeReviewsFromYelp&fakeReviewsFromTemporalAlgo)/len(fakeReviewsFromTemporalAlgo)
precisionOfLBP = len(fakeReviewsFromYelp&fakeReviewsInParentLBP)/len(fakeReviewsInParentLBP)
recallOfTemporalAlgo = len(fakeReviewsFromYelp&fakeReviewsFromTemporalAlgo)/len(fakeReviewsFromYelp)
recallOfLBP = len(fakeReviewsFromYelp&fakeReviewsInParentLBP)/len(fakeReviewsFromYelp)
F1ScoreOfTemporalAlgo = (2*precisionOfTemporalAlgo*recallOfTemporalAlgo)/(precisionOfTemporalAlgo+recallOfTemporalAlgo)
F1ScoreOfLBP = (2*precisionOfLBP*recallOfLBP)/(precisionOfLBP+recallOfLBP)
print 'Accuracy of Temporal LBP',accuracyOfTemporalAlgo
print 'Accuracy of LBP', accuracyOfLBP
print 'Precision of Temporal LBP',precisionOfTemporalAlgo
print 'Precision of LBP', precisionOfLBP
print 'Recall of Temporal LBP', recallOfTemporalAlgo
print 'Recall of LBP', recallOfLBP
print 'F1Score of Temporal LBP',F1ScoreOfTemporalAlgo
print 'F1Score of LBP',F1ScoreOfLBP
def mergeTimeBasedGraphsWithNotMergeableIds(alltimeD_access_merge_graph,not_mergeable_businessids, cross_time_graphs):
# whatever businesses did drastically change,
# we will copy the super graph and try adding these edges to the copied
# graph and run LBP
to_be_removed_edge_between_user_bnss = set()
#copy_merge_lbp_runner_threads = []
beforeThreadTime = datetime.now()
for time_key in not_mergeable_businessids:
copied_all_timeD_access_merge_graph = deepcopy(alltimeD_access_merge_graph)
graph = cross_time_graphs[time_key]
for bnssid in not_mergeable_businessids[time_key]:
bnss = graph.getBusiness(bnssid)
usrs = graph.neighbors(bnss)
for usr in usrs:
review = deepcopy(graph.get_edge_data(usr,bnss)[SIAUtil.REVIEW_EDGE_DICT_CONST])
copied_all_timeD_access_merge_graph.add_edge(copied_all_timeD_access_merge_graph.getBusiness(bnss.getId()),\
copied_all_timeD_access_merge_graph.getUser(usr.getId()),
{SIAUtil.REVIEW_EDGE_DICT_CONST:review})
copy_merge_lbp = LBP(copied_all_timeD_access_merge_graph)
copy_merge_lbp.doBeliefPropagationIterative(50)
(fakeUsers, honestUsers,unclassifiedUsers,\
badProducts,goodProducts, unclassifiedProducts,\
fakeReviewEdges, realReviewEdges,unclassifiedReviewEdges) = copy_merge_lbp.calculateBeliefVals()
for edge in fakeReviewEdges:
(s1,s2) = edge
if s1.getNodeType() == SIAUtil.USER:
to_be_removed_edge_between_user_bnss.add((s1.getId(),s2.getId()))
else:
to_be_removed_edge_between_user_bnss.add((s2.getId(),s1.getId()))
# copy_merge_lbp_runner = LBPRunnerThread(copied_all_timeD_access_merge_graph, 25, 'LBP Runner For Not mergeableIds'+str(time_key))
# copy_merge_lbp_runner_threads.append(copy_merge_lbp_runner)
# copy_merge_lbp_runner.start()
# for copy_merge_lbp_runner in copy_merge_lbp_runner_threads:
# copy_merge_lbp_runner.join()
afterThreadTime = datetime.now()
print 'Time to be reduced', afterThreadTime-beforeThreadTime
# for copy_merge_lbp_runner in copy_merge_lbp_runner_threads:
# print 'Copy merge runner', len(copy_merge_lbp_runner.getFakeEdgesData())
# to_be_removed_edge_between_user_bnss = to_be_removed_edge_between_user_bnss.union(copy_merge_lbp_runner.getFakeEdgesData())
#from the drastically change businesses we have find out all fake edges in the above step
# without them add rest of the edges to the super graph and run LBP on it
for time_key in not_mergeable_businessids:
graph = cross_time_graphs[time_key]
for bnssid in not_mergeable_businessids[time_key]:
bnss = graph.getBusiness(bnssid)
usrs = graph.neighbors(bnss)
for usr in usrs:
if (usr.getId(),bnss.getId()) not in to_be_removed_edge_between_user_bnss:
review = deepcopy(graph.get_edge_data(usr,bnss)[SIAUtil.REVIEW_EDGE_DICT_CONST])
alltimeD_access_merge_graph.add_edge(alltimeD_access_merge_graph.getBusiness(bnss.getId()),\
alltimeD_access_merge_graph.getUser(usr.getId()),\
{SIAUtil.REVIEW_EDGE_DICT_CONST:review})
print "------------------------------------Running Final Merge LBP--------------------------------------"
merge_lbp = LBP(alltimeD_access_merge_graph)
merge_lbp.doBeliefPropagationIterative(50)
(fakeUsers, honestUsers,unclassifiedUsers,\
badProducts,goodProducts, unclassifiedProducts,\
fakeReviewEdges, realReviewEdges,unclassifiedReviewEdges) = merge_lbp.calculateBeliefVals()
for edge in fakeReviewEdges:
to_be_removed_edge_between_user_bnss.add((merge_lbp.getEdgeDataForNodes(*edge).getUserId(),\
merge_lbp.getEdgeDataForNodes(*edge).getBusinessID()))
certifiedRealFromTemporalAlgo = set()
for edge in realReviewEdges:
certifiedRealFromTemporalAlgo.add((merge_lbp.getEdgeDataForNodes(*edge).getUserId(),\
merge_lbp.getEdgeDataForNodes(*edge).getBusinessID()))
return (to_be_removed_edge_between_user_bnss,certifiedRealFromTemporalAlgo)