[PPT] - Social/Network/Analysis mohamed.bouguessa@uqo.ca/ 1 Web/today 2 PowerPoint Presentation

SLIDE 1

1

mohamed.bouguessa@uqo.ca/

Social/Network/Analysis

SLIDE 2

2

Web/today

SLIDE 3

3

Web/today/–/Diverse/applications

SLIDE 4

4

Web/today/–/Millions/of/users

SLIDE 5

5

Web/today/–/Rich/content

SLIDE 6

6

Web/today/–/Highly/dynamic/

SLIDE 7

7

Web/today/–/Traces/of/activity

SLIDE 8

8

Web/today/–/Rich/interactions

Rich/interactions/ between/users/ and/content/

SLIDE 9

9 9

Web/today/–/Interaction/networks

Rich interactions between users and content Modeled as interaction network

SLIDE 10

10

We#can#all#be#connected#through#a#series#of#six#contacts# appeals#to#me.#It#makes#the#world#seem#less#brutal,#and# more#warm#and#more#friendly.##

Six/degrees/of/separation

SLIDE 11

11

Six/degrees/of/separation

SLIDE 12

12

Testing/the/smallGworld/hypothesis

MSN Messenger Average path length is 6.6 90% of nodes is reachable <8 steps Network of who talks to whom on MSN Messenger: 240M nodes, 1.3 billion edges

SLIDE 13

13

Why/study/networks?

Build/understanding/and/theory:/

– How#users#create#content#and#interact#with#it#and# among#themselves?#

Build/better/onGline/applications:/

– How#to#design#better#services#and#algorithms?#

SLIDE 14

14

#A#social/network/is#a#social#structure#of#people,#related#

(directly#or#indirectly)#to#each#other#through#a#common# relation#or#interest.#

/Social/network/analysis/(SNA)/is#the#study#of#social#

networks#to#understand#their#structure#and#behavior.#

Social/Networks/Analysis

SLIDE 15

15

Social/Networks

Social#network:#relationship#among#

interacting#units.#

SLIDE 16

16

Social/Networks

Interacting unites: Actors / nodes discrete individual, corporate, or collective social units

SLIDE 17

17

Relational ties between actors are channels to transfer, exchange or flow of resources. Relations, linkages or ties

Social/Networks

SLIDE 18

18

Social/Networks

Social#network#representation#

– Adjacency#matrix##(socioGmatrix)# – Graph#(SocioGgraph)#

! ! ! ! ! ! ! ! ! ! ! ! " # $ $ $ $ $ $ $ $ $ $ $ $ % & 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

[ ]

9 8 7 6 5 4 3 2 1

! ! ! ! ! ! ! ! ! ! ! ! " # $ $ $ $ $ $ $ $ $ $ $ $ % & 9 8 7 6 5 4 3 2 1

SLIDE 19

19

#Computer#Science#has#created#the#cyber#infrastructure#for#

–#Social#Interaction# –#Knowledge#Exchange# –#Knowledge#Discovery#

#Ability#to#capture#

–#different#about#various#types#of#social#interactions# –#at#a#very#fine#granularity# –#with#practically#no#reporting#bias#

Data/mining/techniques/can/be/used/for/building/ descriptive/and/predictive/models/of/social/interactions/

Key/Drivers/for/CS/Research/in/SNA

SLIDE 20

20

SNA/Techniques

Prominent/problems/

Social#network#extraction/construction#
Identifying#prominent/trusted/expert#actors#
Identifying#Spammers###
Discovering#communities#in#social#networks#
Link#prediction#
Approximating#large#social#networks#
Evolution#of#social#networks#

SLIDE 21

21

#Mining#a#social#network#from#data#sources#
#Recent#research#suggest#that#there#are#three#sources#of#social#

network#data#on#the#web#

#Content#available#on#web#pages#(e.g.#user#homepages,#message#

threads#etc.)#

#User#interaction#logs#(e.g.#email#and#messenger#chat#logs)#
#Social#interaction#information#provided#by#users#(e.g.#social#

network#service#websites#such#as#Orkut,#Friendster#and#MySpace)#

W e b D

c

u m e n t s C

m

m u n i c a t i

n

L

g

s P r

f

i l e _ 3 P r

f

i l e _ 1 P r

f

i l e _ 5 P r

f

i l e _ 4 P r

f

i l e _ 2 A c t

r

p r

f

i l e s

n

a S

c

i a l N e t w

r

k S e r v i c e W e b D

c

u m e n t s C

m

m u n i c a t i

n

L

g

s C

m

m u n i c a t i

n

L

g

s P r

f

i l e _ 3 P r

f

i l e _ 1 P r

f

i l e _ 5 P r

f

i l e _ 4 P r

f

i l e _ 2 A c t

r

p r

f

i l e s

n

a S

c

i a l N e t w

r

k S e r v i c e P r

f

i l e _ 3 P r

f

i l e _ 1 P r

f

i l e _ 5 P r

f

i l e _ 4 P r

f

i l e _ 2 A c t

r

p r

f

i l e s

n

a S

c

i a l N e t w

r

k S e r v i c e

Social/Network/Extraction

SLIDE 22

22

Social/Network/Extraction

Extracting#a#social#network#

– Asking#people#about#their#relations# – Tracking#their#contacts#(emails,#phone#call,# visits,#etc.)#such#as#Enron#project# – Mining#their#contextual#data#(papers,# interviews,#resumes,#news,##biographies,# citations,#references,#web#pages,#blogs,# portfolios,#etc.)#!#Learning#social#network#

SLIDE 23

23

Learning/Social/Networks

Learning##social#network#from#text#

– Descriptive#vs.#Predictive#model# – We#only#predict#the#possible#relations# between#the#actors#

SLIDE 24

24

Learning/Social/Networks

Usually,#we#can#reach#documents#by# knowing#people…#

#

SLIDE 25

25

Learning/Social/Networks

…and#directly#or#indirectly#we#will#know#

ther#documents#by#(or#about)#other#people#

through#these#documents…# # #

SLIDE 26

26

…and#very#soon#we#will#have#a#social#network# including#some#individuals#who#have#been# connected#to#each#other#via#some#similar# contents.# #

Learning/Social/Networks

SLIDE 27

27

SNA/Techniques

Prominent/problems/

Social#network#extraction/construction#
Identifying/prominent/trusted/expert/actors/
Identifying#Spammers###
Discovering#communities#in#social#networks#
Link#prediction#
Approximating#large#social#networks#
Evolution#of#social#networks#

SLIDE 28

28

Identifying/prominent/expert/actors/ in/social/networks/

Link/Analysis/Technique/

HITS/
PageRank/

SLIDE 29

29

#Being#Authority#depends#upon#inGedges;#an#authority#has#a#large#

number#of#edges#pointing#towards#it.#

#Being#a#Hub#depends#upon#outGedges;#a#hub#links#to#a#large#

number#of#nodes.#

#Notice#that#the#definition#of#hubs#and#authorities#is#circular.#
#Nodes#can#be#both#hubs#and#authorities#at#the#same#time#

Hubs/and/Authorities

SLIDE 30

30

∑

→

=

p q

q h p a ) ( ) (

∑

→

=

q p

q a p h ) ( ) (

Hubs/and/Authorities

SLIDE 31

31

# The# PageRank# assumption# is# that# a# node# transfers# its#

PageRank#values#evenly#to#all#the#nodes#it#connects#to.##

##A#node#has#high#rank#if#the#sum#of#the#ranks#of#its#inGlinks#is#

high.###

#This#covers#both#the#case#where#a#node#has#many#inGlinks#

and#that#where#a#node#has#a#few#highly#ranked#inGlinks.#

Google’s/PageRank

SLIDE 32

32

How/is/PageRank/calculated?/

! " # $ % & + + + − = ) ( ) ( ) 1 ( ) 1 ( * ) 1 ( ) ( Tn C Tn PR T C T PR d d A PR 

C(Ti):#the#number#of#OutGlinks#of#the#page/node#Ti#

That's#the#equation#that#calculates#a#page's#PageRank.#It's#the#

riginal#one#that#was#published#when#PageRank#was#being#

developed,#and#it#is#probable#that#Google#uses#a#variation#of#it# but#they#aren't#telling#us#what#it#is.#It#doesn't#matter#though,#as# this#equation#is#good#enough.##

Google’s/PageRank

SLIDE 33

33

PR(A)#=#PR(B)#=#PR(C)#=PR(D)#=1# PR(A)#>#PR(B)#>#PR(C)#>#PR(D)##

Google’s/PageRank

SLIDE 34

34

A B C D A B C D E A B C D E A B C D E

Google’s/PageRank

SLIDE 35

35

Yahoo!/Answers/:/Identifying/the/expert/

User_x User_y User_z User Votes

Question

Answer_1

User_1

Answer_2

User_2

Answer_n

User_n

#Identifying#the#true#experts#among#Yahoo#Answers#participants#
#Keep#track#of#users#who#consistently#provide#good#answers#for#particular#topics#
#Provide#incentives#for#experts#to#stay#on#Yahoo!#Answers#in#order#to#improve#

service#

SLIDE 36

36

Yahoo!/Answers

SLIDE 37

37 37

Question/Life/Cycle

SLIDE 38

38

3 4 1 2 1 4 1 2 1 3 2 4 1 2 1 3 5

Users who usually only ask questions Users who usually only answer questions Users who help each other

Example#of#interactions#between#askers#and#best#answerers##

Yahoo!/Answers

How$to$es(mate$the$authority$degree$for$each$user?

SLIDE 39

39

PageRank?

Example:#The#category#of#“Programming”/

#User#B#answers#user#A’s#questions,#which#are#about#Java;#
#User#C#answers#B’s#questions,#which#are#about#PHP;###

" #Is#it#possible#to#state#that#C#is#more#expert#than#B?##

#No,#because:#B#and#C#have#different#expertise.##

A B C

JAVA PHP

SLIDE 40

40

HITS?

#The#HITS#algorithm#assigns#high#authority#scores#to#

nodes#1,#2,#10,#11#and#12,#but#a#nearGzero#authority#score# to#node##

SLIDE 41

41

HITS?

The#HITS#algorithm#will#allocate#high#authority#scores#to#the#nodes#N9–N15,#

while#giving#zero#authority#score#to#node#N1.##

The#reason#for#this#is#quite#similar#to#example#1.##
Specifically,#the#fact#that#node#N8#points#to#many#nodes#contribute#to#increase#

its#hub#score.#Hence,#causing#the#nodes#N9–N15#to#receive#higher#authority# scores.#However,#intuition#suggests#that#node#N1#is#the#most#authoritative# since#it#represents#an#answerer#with#a#large#number#of#best#answers.#

SLIDE 42

42

Proposed/Approach

#The#authority#score#of#each#user#is#simply#the#number#of#best#

answer#of#each#users#normalized#so#their#square#sum#to#1:##

∑ =

=

N i i

y

1 2

1 ) (

$yi#provide#a#relative#score#of#the#authority#of#each#user#in#each#

category.##

" #We#are#interested#in#all#sets#of#Ui#having#large#values#of#yi.###

SLIDE 43

43

Authority/Score

#Example:#Category#of#“Engineering”#

SLIDE 44

44

Authority/Score

SLIDE 45

45

Automatic/Identification/of/Authorities

SLIDE 46

46

Experiments

We#conduct#experiments#on#datasets#which#represent#users’# activities#over#one#full#year#for#six#categories:##

Category % users who ask only % users who answer only % users who ask and answer Engineering 65% 31% 4% Biology 60% 36% 4% Programming 66% 29% 5% Mathematics 64% 31% 5% Physics 60% 34% 6% Chemistry 63% 32% 5%

SLIDE 47

47

Authoritative/Users

SLIDE 48

48

Quality/of/Content

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Biology Chemistry Engineering Mathematics Physics Programming Category Average quality score

#The#identified#authoritative#users#generate#highGquality#

content#in#Yahoo!#Answers.##

#Askers#are#very#selective#in#choosing#the#best#answerers#

SLIDE 49

49

SNA/Techniques

Prominent/problems/

Social#network#extraction/construction#
Identifying#prominent/trusted/expert#actors#
Identifying/Spammers///
Discovering#communities#in#social#networks#
Link#prediction#
Approximating#large#social#networks#
Evolution#of#social#networks#

SLIDE 50

50

Existing/Approaches

Use#supervised#machine#learning#methods#to#

perform#spam#classification.#

1. First,#extract#several#features#that#

characterize#the#activities#of#each#user#in#the# network.#

2. Then,#a#supervised#learning#model#is#used#to#

learn#the#behavior#of#spammers#and# legitimate#users.#

3. Finally,#the#trained#model#is#applied#to#filter#
ut#spammers#

SLIDE 51

51

Why/new/approach?

" Existing#approaches#suffer#from#their# dependency#on#the#training#data.# " Robust#learning#approach#for#spammer# detection#often#requires#large#amounts#of# labeled#training#data.# " The#problem#here#is#that#labeled#samples# are#more#difficult,#expensive#and#time# consuming#to#obtain#than#unlabeled#ones.#

SLIDE 52

52

Why/new/approach?

The#scarcity#of#trained#data#is#one#of#the#major#

challenges#facing#current#spammer#detection# methods#in#social#networks.#

Unlabeled#data#are#relatively#easy#to#collect#

" #There#is#good#reason#to#focus#on# unsupervised#approaches#to#explore#the# vast#amount#of#unlabeled#data#available#at# low#cost.#

SLIDE 53

53

Proposed/Approach

We#investigate#the#link#structure#of#the#network#

to#discriminate#between#spammer#and# legitimate#users.#

1. First,#we#estimate#a#user’s#reputation#level#

score#within#a#network#

2. Then,#we#propose#a#probabilistic#approach#

based#on#the#beta#mixture#model#to#identify# spammers##

SLIDE 54

54

Interactions/in/Social/Networks

Interactions#between#normal#users#share#a#

similar#pattern#such#that#when#a#user#receives#a# message#from#a#legitimate#sender,#he/she#will,# generally,#reply#to#this#message.#

spammers#tend#to#send#messages#in#bulk#and#

receive#few#or#no#replies#back#from#their# recipients.#

SLIDE 55

55

Communication/Reciprocity

xi$communication#reciprocity#of#a#user#ui$
OSi$is#a#set#of#accounts#that#received#at#least#one#

message#from#a#user#ui$

ISi$is#a#set#of#accounts#that#sent#at#least#one#

message#to#ui.#

SLIDE 56

56

Communication/Reciprocity

CR#measures#the#probability#of#a#node#receiving#a#

response#from#each#of#its#neighbors.# " Legitimate#accounts#have#a#much#higher# probability#of#being#responded#to.# " spammers#have#a#very#low#response#rate## # CR#value#close#to#zero#

SLIDE 57

57

The/Probabilistic/Model

The#estimated#CR#scores#can#be#considered#as#

coming#from#several#underlying#probability# distributions.#

Each#distribution#is#a#component#of#the#

mixture#model#representing#users#with#close# CR#values#

All#the#components#are#combined#into#a#

comprehensive#model#by#a#mixture#form.#

SLIDE 58

58

We#propose#to#use#the#beta#mixture#

model#due#to#its#shape#flexibility#

The/Probabilistic/Model

SLIDE 59

59

The/Probabilistic/Model

Formally,#we#expect#that#the#CR#scores#follow#

a#mixture#density#of#the#form#

The#density#function#of#the#lth#component#is#

given#by#

SLIDE 60

60

The/Probabilistic/Model

#How#to#estimate#the#parameters#of#the#beta#

components?# Maximum#likelihood#technique#

SLIDE 61

61

The/Probabilistic/Model

#How#to#estimate#the#number#of#components?#
1. Calculate#the#maximum#likelihood#of#the#

parameters#of#the#mixture#for#a#range#of#values#

f#m#(from#1#to#m_max).#
2. Calculate#an#associated#criterion#and#selecting#

the#value#of#m#which#optimizes#the#criterion.## ## ##

SLIDE 62

62

The/Probabilistic/Model

#How#to#estimate#the#number#of#components?#

Integrated/Classification/Likelihood/// Bayesian/Information/Criteria/(ICLGBIC)/

SLIDE 63

63

Procedure

SLIDE 64

64

Experiments/:/Data/Sets

We#employ#a#data#set#representing#the#email#

network#of#the#University#at#Rovira#i#Virgili#(URV)#in# Tarragona,#Spain# " 1,133#users#(faculty,#reserachers,#etc.)# " all#legitimate#users# " the#URV#email#network#icludes#no#spam#sender# " Transactions/from/spammers/are/therefore/ simulated/by/generating/mock/spam/accounts/ in/the/URV/email/network/data/

SLIDE 65

65

Data/Sets

We#generated#six#different#sets#of#simulated#

spam#accounts#in#order#to#inject#spam#traffic#into# the#URV#data#set#of#legitimate#senders.## " Data200,#Data400,#Data600,#Data800,# Data1000,#and#Data1200,#containing#200,#400,# 600,#800,#1,000#and#1,200#spam#accounts# respectively,#

number#of#legitimate#senders#was#1,133#in#

each#set.#

SLIDE 66

66

Comparison

We#compare#our#approach#to#that#of#MailNet#a#

supervised#learning#approach#for#detecting# spammers#on#email#social#networks.## " MailNet#first#extracts#a#number#of#features#from# each#user#in#the#email#social#network.## " Then,#support#vector#machine#learning#is#used# to#perform#spam#classification#

SLIDE 67

67

Results

The#first#component#represents#spam#accounts#with#small#CR#values#

which#are#close#to#zero.##

The#third#component#corresponds#to#accounts#with#CR#scores#close#to#
ne#(users#who#get#replies#to#the#vast#majority#or#all#of#the#messages#

that#they#sent).##

The#second#component#corresponds#to#users#who#receive,#in#general#

replies#to#most#of#the#messages#they#send#

SLIDE 68

68

Results

SLIDE 69

69

Results

Our#unsupervised#method#performs#as#well#as#

the#SVMGbased#approach#MailNet.#

But#our#approach#also#has#the#ability#to#mine#

unlabeled#data#G#a#considerable#practical# advantage#for#realGworld#applications#in# which#class#labels#are#not#available.#

SLIDE 70

70

Application/to/Yahoo!/Answers

SLIDE 71

71 71

Question/Life/Cycle

SLIDE 72

72

Yahoo!#Answers#also#has#a#social#networking#

capability#where#users#can#connect#and#make# friends#with#other#members#who#share#similar# interests.#

In#our#experiments,#we#analyzed#a#network#

consisting#of#167,455#Yahoo!#Answers#users# accounts,#using#our#approach#to#identify# spammers#

Yahoo!/Answers/Social/Network/

SLIDE 73

73

Results

" #we#found#that#32,724#accounts#could#be# classified#as#spam.#

SLIDE 74

74

Quality/of/Content

We#evaluated#the#quality#of#content#generated#

by#users#identified#as#spammers#and#compared# it#to#the#quality#of#content#generated#by#users# classified#as#nonGspammers.#

We#expect#that#normal#users#will#generate#

relatively#goodGquality#content##

SLIDE 75

75

In# order# to# estimate# the# quality# of# the# content#

generated#by#the#identified#authoritative#users#we#use# the#algorithm#of#Agichtein#et#al.#(WSDM#2008).#

In# brief:#The# approach# combines# the# analysis# of# the#

textual#content#with#the#user#feedback#on#the#site#in#

rder#to#estimate#a#quality#score#for#each#question#and#

answer.##

The#quality#score#is#the#confidence#score#of#a#binary#

classifier#trained#on#high#and#low#quality#examples.##

When# the# question# or# answer# is# of# high# quality,# the#

value#of#the#quality#score#is#close#to#1##

Quality/of/Content

SLIDE 76

76

Quality/of/Content

legitimate#users#generate#relatively#better##

quality#content#than#do#spammers.#

SLIDE 77

77

SNA/Techniques

Prominent/problems/

Social#network#extraction/construction#
Identifying#prominent/trusted/expert#actors#
Identifying#Spammers###
Discovering/communities/in/social/networks/
Link#prediction#
Approximating#large#social#networks#
Evolution#of#social#networks#

SLIDE 78

78

Community/Structure///// in/Social/Network

Non-Sybil Region Sybil Region

SLIDE 79

79

Graph/Clustering

SLIDE 80

80

Algorithms+based+on+Czekanovski5Dice+Distance

( ) ( ) ( ) ( )

2 1 2 1 2 1 2 1 ) 2 , 1 ( S S S S S S S S N N dist     + − =

Distance#between#two#nodes#

S1:#number#of#nodes#connected#to#N1#(including#N1)# S2:#number#of#nodes#connected#to#N2#(including#N2)# # Small#distance#$#High#similarity#

SLIDE 81

81

Czekanovski5Dice+Distance

N1 N2 N3 N4 N5 N6

/Exemple/
#dist(N1,#N2)#=#?##

S1#=#{N1,#N2,#N3}# S2#=#{N2,#N1,#N3}#

( ) ( ) ( ) ( )

3 3 3 3 2 1 2 1 2 1 2 1 ) 2 , 1 ( = + − = + − = S S S S S S S S N N dist    

#dist(N3,#N4)#=#?##

S3#=#{N3,#N1,#N2,#N4}# S4#=#{N4,#N3,#N5,#N6}#

( ) ( ) ( ) ( )

5 . 2 6 2 6 4 3 4 3 4 3 4 3 ) 4 , 3 ( = + − = + − = S S S S S S S S N N dist    

SLIDE 82

82

Czekanovski5Dice+Distance

82

(a)#Graph# (b)#Smilarity# Matrix# (c)#Dendogramme# (d)#Clustering#

SLIDE 83

83

Applica9on

The#Santa#Fe#Institute#collaboration#network#

SLIDE 84

84

Applica9on

Enron#email#network##

SLIDE 85

85

Discovering+Knowledge5Sharing+ Communi9es+in+ + Ques9on5Answering+Forums

SLIDE 86

86

Knowledge5Sharing+Community

1. A#knowledgeGsharing#community#is#defined#by#a#

set#of#askers#and#authoritative#users.#

2. Within#each#community,#askers#exhibit#more#

homogenous#behavior#in#terms#of#their# interactions#with#authoritative#users#than# elsewhere.#

3. Authoritative#users#may#belong#to#more#than#one#

community.#

SLIDE 87

87

Knowledge5Sharing+Community

Existing#graphGbased#community#detection# methods#are#not#appropriate#for#our#study.##

SLIDE 88

88

Example

a1 : e1, e2 a2 : e1, e2 a3 : e2, e3 a4 : e2, e3 a5 : e1, e2, e3 a6 : e1, e2, e3

SLIDE 89

89

Example

Modeling#users#interactions#as#a#graph##

SLIDE 90

90

The+GRACLUS+Algorithm

SLIDE 91

91

Modeling+Interac9ons+Between+Users

" #We#use#a#transactional#data#model#to#represent#the# interactions#between#askers#and#authoritative#users.#

#The#first#community#is#defined#by#T1,$T2,$T3#et#T4$
$The#second#community#is#defined#by#T5,$T6,$T7$et$T8$

SLIDE 92

92

#Boolean#representation#of#the#interaction#between# askers#and#authoritative#users.#

Illustra9on

SLIDE 93

93

The+TRANCLUS+Algorithm

#A$=$${a1,$a2,$…,$an}#a#set#of#n#askers#
#E$=${e1,$e2,$…,$ed}#a#set#of#d#authoritative#users#
#TD$={T1,$T2,$…,$Tn}#a#collection#of#n#transactions#that#

summarizes#the#interactions#of#all#askers#ai$with#the# identified#authoritative#users.#

SLIDE 94

94

Problem+Defini9on

Given#the#set#A#of#askers#and#the#set#E#of# authoritative#users,##

#Construct#the#set#TD.#
#Partition#TD#into#a#set#of#disjoint#clusters############

###C=${C1,$C2,$…,$Cnc}## " #The#identified#clusters#represent#the#communities# we#want#to#discover.#

SLIDE 95

95

Criterion+Func9on

( )

∑ ∑

= ∈

# $ % & ' ( × =

nc s C e s s

s

e Z C e

cc

n n C CF

1 3 2

) ( ) , ( 1 1 ) (

( )

1 ) , ( ) ( + − = TD e

cc

n e Z

SLIDE 96

96

The+TRANCLUS+Scheme

SLIDE 97

97

Applica9on+to+Yahoo!+Answers

SLIDE 98

98

Content+Analysis

" #The#clustered#askers#tend#to#post#questions#on#closed# related#topics##

SLIDE 99

99

Influence/of/Social/Networks/on/Product/Recommendations/

A1 A2 . . . AN A1 A2 A3 … AN P1 P2 P3 … PM A1 A2 . . . AN

Recommendation System

Social Network Product Opinion

#Understanding#the#impact#of#social#networks#on#market#behavior#
#Improved#recommendation#systems#

Emerging+Applica9on

SLIDE 100

100

Tha hat’s t’s all fol

lks!