Data Mining for Social Network Analysis Australasian Data Mining - - PowerPoint PPT Presentation

▶
data mining for social network analysis
SMART_READER_LITE
LIVE PREVIEW

Data Mining for Social Network Analysis Australasian Data Mining - - PowerPoint PPT Presentation

Jan 27, 2023 22 likes •774 views

Data Mining for Social Network Analysis Australasian Data Mining Conference (AusDM) 2007 December 3 rd 4 th , 2007 Jaideep Srivastava University of Minnesota Gold Coast, Australia srivasta@cs.umn.edu Joint work with: Arindam Banerjee,

slide-1
SLIDE 1

12/2/2007 1

Data Mining for Social Network Analysis

Jaideep Srivastava University of Minnesota srivasta@cs.umn.edu

Joint work with: Arindam Banerjee, Nishith Pathak, Sandeep Mane, Muhammad A. Ahmad David Kuo-Wei Hsu, Young Ae Kim, University of Minnesota Noshir S. Contractor, Northwestern University Dmitri Williams, University of Southern California Sony Online Entertainment Special thanks to Enron (via US DoJ) Sponsors: US National Science Foundation US Army Research Institute Digital Technology Center, University of Minnesota

Australasian Data Mining Conference (AusDM) 2007 December 3rd – 4th, 2007 Gold Coast, Australia

slide-2
SLIDE 2

12/2/2007 Jaideep Srivastava 2

  • Introduction to Social Network Analysis (SNA)
  • Computer Science and SNA
  • A Detailed Case Study
  • Socio-cognitive analysis from e-mail logs

Modeling socio-cognitive networks Analysis of a socio-cognitive network Experiments with the Enron dataset

  • Extracting concealed relationships

An IR-inspired approach

  • Other applications
  • SNA from MMORPG logs
  • Trust in social networks
  • Expert finding in social networks
  • Social networks in health care management
  • Some Emerging Applications
  • References

Outline

slide-3
SLIDE 3

Introduction to Social Network Analysis Introduction to Social Network Analysis

slide-4
SLIDE 4

12/2/2007 Jaideep Srivastava 4

Social Networks

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

slide-5
SLIDE 5

12/2/2007 Jaideep Srivastava 5

SNA in Popular Science Press

Social Networks have captured the public imagination in recent years as evident in the number of popular science treatment of the subject

slide-6
SLIDE 6

12/2/2007 Jaideep Srivastava 6

Types of Networks (Contractor, 2006)

Social Networks

“who knows who”

Socio-Cognitive Networks

“who thinks who knows who”

Knowledge Networks

“who knows what”

Cognitive Knowledge Networks

“who thinks who knows what”

Networks in Social Sciences

slide-7
SLIDE 7

12/2/2007 Jaideep Srivastava 7

Types of Social Network Analysis

Sociocentric (whole) network analysis

Emerged in sociology Involves quantification of interaction among a socially well- defined group of people Focus on identifying global structural patterns Most SNA research in organizations concentrates on sociometric approach

Egocentric (personal) network analysis

Emerged in anthropology and psychology Involves quantification of interactions between an individual (called ego) and all other persons (called alters) related (directly

  • r indirectly) to ego

Make generalizations of features found in personal networks Difficult to collect data, so till now studies have been rare

slide-8
SLIDE 8

12/2/2007 Jaideep Srivastava 8

Networks Research in Social Sciences

Social science networks have widespread application in various fields Most of the analyses techniques have come from Sociology, Statistics and Mathematics See (Wasserman and Faust, 1994) for a comprehensive introduction to social network analysis

Epidemiology Sociology Organizational Theory Social Psychology Anthropology

Socio-Cognitive Networks Cognitive Knowledge Networks Social Networks Knowledge Networks Perception Reality Acquaintance (links) Knowledge (content)

Epidemiology Sociology Organizational Theory Social Psychology Anthropology

Socio-Cognitive Networks Cognitive Knowledge Networks Social Networks Knowledge Networks Perception Reality Acquaintance (links) Knowledge (content) Socio-Cognitive Networks Cognitive Knowledge Networks Social Networks Knowledge Networks Socio-Cognitive Networks Cognitive Knowledge Networks Social Networks Knowledge Networks Perception Reality Acquaintance (links) Knowledge (content)

slide-9
SLIDE 9

Computer Science and Social Network Analysis

slide-10
SLIDE 10

12/2/2007 Jaideep Srivastava 10

Computer networks as social networks

“Computer networks are inherently social networks, linking people, organizations, and knowledge” (Wellman, 2001) Data sources include newsgroups like USENET; instant messenger logs like AIM; e-mail messages; social networks like Orkut and Yahoo groups; weblogs like Blogger; and online gaming communities

slide-11
SLIDE 11

12/2/2007 Jaideep Srivastava 11

Key Drivers for CS Research in SNA

Computer Science has created the über-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

  • Fertile research area for data mining research
slide-12
SLIDE 12

12/2/2007 Jaideep Srivastava 12

A shift in approach: from ‘synthesis’ to ‘analysis’

  • !

! " # $ $!

  • %

" &' (#$ )

  • *!
  • (+

, (+

  • (+

#! !

  • *
  • ,

! Shift in approach

slide-13
SLIDE 13

Data Mining for SNA Case Study

Socio-Cognitive Analysis from E-mail Logs

slide-14
SLIDE 14

12/2/2007 14

Modeling a Socio-Cognitive Network

slide-15
SLIDE 15

12/2/2007 Jaideep Srivastava 15

Example of E-mail Communication

A sends an e-mail to B

With Cc to C And Bcc to D

C forwards this e-mail to E From analyzing the header, we can infer

A and D know that A, B, C and D know about this e-mail B and C know that A, B and C know about this e-mail C also knows that E knows about this e-mail D also knows that B and C do not know that it knows about this e- mail; and that A knows this fact E knows that A, B and C exchanged this e-mail; and that neither A nor B know that it knows about it and so on and so forth …

slide-16
SLIDE 16

12/2/2007 Jaideep Srivastava 16

Modeling Pair-wise Communication

Modeling pair-wise communication between actors

Consider the pair of actors (Ax,Ay) Communication from Ax to Ay is modeled using the Bernoulli distribution L(x,y)=[p,1-p] Where,

p = (# of emails from Ax with Ay as recipient)/(total # of emails exchanged in the network)

For N actors there are N(N-1) such pairs and therefore N(N-1) Bernoulli distributions Every email is a Bernoulli trial where success for L(x,y) is realized if Ax is the sender and Ay is a recipient

slide-17
SLIDE 17

12/2/2007 Jaideep Srivastava 17

Modeling an agent’s belief about global communication

Based on its observations, each actor entertains certain beliefs about the communication strength between all actors in the network A belief about the communication expressed by L(x,y) is modeled as the Beta distribution, J(x,y),

  • ver the parameter of L(x,y)

Thus, belief is a probability distribution over all possible communication strengths for a given

  • rdered pair of actors (Ax,Ay)
slide-18
SLIDE 18

12/2/2007 Jaideep Srivastava 18

Model for Belief Update

Jk(x,y) is the Beta distribution maintained by actor Ak regarding its belief about the communication from Ax to Ay a and b, the two parameters of Jk(x,y), are associated with the number of emails observed by Ak which are

from Ax to Ay , i.e. number of successes, and from Ax not to Ay, i.e. number of failures

Initialization

a and b start out with default initial values Many different possibilities

For example, values can be chosen to be small so that they do not have much of an impact and can be “washed out” by future

  • bservations

Belief update

  • n observing a success or failure, Ak increments a or b

respectively

slide-19
SLIDE 19

12/2/2007 Jaideep Srivastava 19

Belief State of an Actor

Every actor maintains Beta distributions (or beliefs) for all

  • rdered pairs of actors in the network

Actor Ak’s belief state is defined to be the set of all N(N-1) Beta distributions (one for every Bernoulli distribution) We also introduce a “super-actor” in the network The super-actor is an actor who observes all the communication in the network Super-actor is used as the baseline for reality E-mail server is the “super-actor”

slide-20
SLIDE 20

12/2/2007 20

Quantitative Measures for Perceptual Closeness

slide-21
SLIDE 21

12/2/2007 Jaideep Srivastava 21

Types of Perceptual Closeness

We analyze the following aspects

Closeness between an actor’s belief and reality, i.e. “true knowledge” of an actor Closeness between the beliefs of two actors, i.e. the “agreement” between two actors

We define two metrics, r-closeness and a- closeness for measuring the closeness to reality and closeness in the belief states of two actors respectively

slide-22
SLIDE 22

12/2/2007 Jaideep Srivastava 22

Measuring the Closeness Between Beliefs

For measuring the closeness between two belief states, the KL-divergence across the expected Bernoulli distributions for the two respective beliefs is computed.

The expected Bernoulli distribution for a belief is the expectation of the Beta distribution corresponding to that belief If J(a,b)k,t is the Beta distribution, then the corresponding expected Bernoulli distribution (denoted by E[J(a,b)k,t]) is obtained by normalizing the parameters of Beta distribution J(a,b)k,t

slide-23
SLIDE 23

12/2/2007 Jaideep Srivastava 23

Belief Divergence Measures

The divergence of one belief, expressed by the Beta distribution J(a,b)x,t, from another, expressed by J(a,b)y,t at a given time t, is defined as, where, and The divergence of a belief state By,t from the belief state Bx,t for two actors Ay and Ax respectively, at a given time t, is defined as,

slide-24
SLIDE 24

12/2/2007 Jaideep Srivastava 24

Belief Divergence Measures (contd.)

The a-closeness measure is defined as the level of agreement between two given actors Ax and Ay with belief states Bx,t and By,t respectively, at a given time t and is given by, The r-closeness measure is defined as the closeness of the given actor Ak’s belief state Bk,t to reality at a given time t and it is given by, Where BS,t is the belief state of the super-actor AS at time t

slide-25
SLIDE 25

12/2/2007 Jaideep Srivastava 25

Interpretation of the Metrics

The r-closeness measure

An actor who has accurate beliefs regarding only few communications is closer to reality than some other actor who has a relatively large number of less accurate beliefs Thus, accuracy of knowledge is important

The a-closeness measure between actor pairs

Consider three actors Ax, Ay and Az Suppose we want to determine how divergent are Ay’s and Az’s belief states from that of Ax’s If Ay and Ax have few beliefs in common, but low divergence for each of these few common beliefs, then their belief states may be closer than those

  • f Az and Ax, who have a relatively larger number of common beliefs with

greater divergence across them

a-closeness measure can be used to construct an “agreement graph” (or a who agrees with whom graph)

Actors are represented as nodes and an edge exists between two actors

  • nly if the agreement or the a-closeness between them exceeds some

threshold t

slide-26
SLIDE 26

12/2/2007 26

r-closeness and a-closeness experiments with Enron E-mail logs

slide-27
SLIDE 27

12/2/2007 Jaideep Srivastava 27

Enron Email Logs

Publicly available: http://www.cs.cmu.edu/~enron/ Cleaned version of data

151 users, mostly senior management of Enron Approximately 200,399 email messages Almost all users use folders to organize their emails The upper bound for number of folders for a user was approximately the log of the number

  • f messages for that user

A visualization of Enron data (Heer, 2005)

For experiments emails exchanged between users for the months of October 2000 and October 2001 were used

slide-28
SLIDE 28

12/2/2007 Jaideep Srivastava 28

Testing ‘conventional wisdom’ using r-closeness

Conventional wisdom 1: As an actor moves higher up the organizational hierarchy, it has a better perception of the social network

It was observed that majority of the top positions were

  • ccupied by employees

Conventional wisdom 2: The more communication an actor observes, the better will be its perception of reality

Even though some actors observed a lot of communication, they were still ranked low in terms of r-closeness. These actors focus on a certain subset of all communications and so their perceptions regarding the social network were skewed towards these “favored” communications Executive management actors who were communicatively active exhibited this “skewed perception” behavior

which explains why they were not ranked higher in the r- closeness measure rankings as expected in 1

slide-29
SLIDE 29

12/2/2007 Jaideep Srivastava 29

Experiment with r-closeness – Oct 2000

For October 2000, based on their r-closeness rankings actors can be roughly divided into three categories

Top ranks: Actors who are communicatively active and observe a lot of diverse communications Mid ranks: Actors who also observe a lot of communication but had skewed perceptions Low ranks: Actors who are communicatively inactive and hardly

  • bserve any of the communication

Ranks Not Avail

  • able

Emplo

  • yees

Higher Manage

  • ment

Executive Manage

  • ment

Others

1-10 2.6% (1) 14.6% (6) 0% (0) 6.9% (2) 6.67% (1) 11–50 28.9 % (11) 34.1% (14) 21.4% (6) 24.1% (7) 13.33% (2) 51-151 68.5% (26) 51.3% (21) 78.6% (22) 69% (20) 80% (12)

Total

100% (38) 100% (41) 100% (28) 100% (29) 100% (15)

slide-30
SLIDE 30

12/2/2007 Jaideep Srivastava 30

Experiment with r-closeness – Oct 2001

r-closeness rankings for the crisis month Oct, 2001 show a significant increase (31% to 65.5%) in the percentage of senior executive management level actors in the top 50 ranks, with employees moving down

Ranks Not Avail- Able Emplo- yees Higher Manage- ment Executive Manage- ment Others

1-10 7.9 % (3) 9.75% (4) 0% (0) 10.3% (3) 0% (0) 11–50 21.1 % (8) 17.1% (7) 25% (7) 55.2% (16) 13.33% (2) 51-151 71% (27) 73.15% (30) 75% (21) 34.5% (10) 86.67% (13)

Total

100% (38) 100% (41) 100% (28) 100% (29) 100% (15)

slide-31
SLIDE 31

12/2/2007 Jaideep Srivastava 31

Agreement Graph for Oct 2000 (threshold = 0.95) Agreement Graph for Oct 2001 (threshold = 0.95)

Mean a-closeness against time

0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 6 _ 1 9 9 9 8 _ 1 9 9 9 1 _ 1 9 9 9 1 2 _ 1 9 9 9 2 _ 2 4 _ 2 6 _ 2 8 _ 2 1 _ 2 1 2 _ 2 2 _ 2 1 4 _ 2 1 6 _ 2 1 8 _ 2 1 1 _ 2 1 1 2 _ 2 1 2 _ 2 2 4 _ 2 2 6 _ 2 2 Time M e a n a

  • c

lo s e n e s s a c ro s s a c to rs

Mean a-closeness against time

Experiment with a-closeness

slide-32
SLIDE 32

12/2/2007 Jaideep Srivastava 32

a-closeness between Lavorato, Denailey, Lay and Skilling

Lavorato Skilling Lay Denailey Lavorato Skilling Lay Denailey Lavorato Skilling Lay Denailey

Initial State up to Oct,1999

0.01 0.01 0.01 0.01 0.01 1.00 Lavorato Skilling Lay Denailey 0.077 0.007 0.019 0.005 0.007 0.016 Lavorato Skilling Lay Denailey 0.06 0.016 0.012 0.03 0.011 0.016 0.047 0.017 0.019 0.027 0.012 0.017 0.049 0.015 0.019 0.027 0.012 0.017

Nov,1999-Dec,2000 Dec, 2000 Jan, 2001- Aug, 2001 Aug, 2001 Sept,2001-Oct,2001 Oct, 2001 Nov,2001- Dec,2001 Dec, 2001

slide-33
SLIDE 33

12/2/2007 33

Automatic Extraction of Concealed Relations

slide-34
SLIDE 34

12/2/2007 Jaideep Srivastava 34

Concealed Relations

Concealed/Covert Relations: Relations between groups of actors that

have high strength but are known to very few actors in the network outside the group

Problem: Given email log data for all actors, extract the concealed relations from this data

slide-35
SLIDE 35

12/2/2007 Jaideep Srivastava 35

An IR-Motivated Approach

Use an approach motivated by informational retrieval Use a tf-idf style scheme for relations

an actor’s view of the social network document in a corpus an (unordered) pair-wise actor relation a term in a document number of instances of a relation observed by an actor term frequency (tf) in a document number of actors that know about a relation document frequency (df) actual frequency of a relation is used to determine a ‘global’ ranking of concealed relations

slide-36
SLIDE 36

12/2/2007 Jaideep Srivastava 36

tf-idf in a Social Network

If N is the total number of actors, then for relation rkl between two actors Ak and Al, we define its tf-idf score to be The score skl induces a ranking among relations based on decreasing levels of concealment If nkl is replaced by mkl

j (# of instances of rkl observed by Aj) then we

get score for relation rkl relative to actor Aj (denoted by skl

j)

The higher the value for skl

j the more privy is actor Aj to the relation

rkl

slide-37
SLIDE 37

12/2/2007 Jaideep Srivastava 37

Top 10 Concealed Relations

Score of this relation has dropped

slide-38
SLIDE 38

12/2/2007 Jaideep Srivastava 38

Top actors from top clusters

slide-39
SLIDE 39

12/2/2007 39

SNA from MMORPG logs

slide-40
SLIDE 40

12/2/2007 Jaideep Srivastava 40

MMO Games

MMO (Massively Multiplayer Online) Games are computer games that allow hundreds to thousands of players to interact and play together in a persistent online world

Popular MMO Games- Everquest 2, World of Warcraft and Second Life

slide-41
SLIDE 41

12/2/2007 Jaideep Srivastava 41

Examples MMOs http://everquest.station.sony.com/ http://www.worldofwarcraft.com/burningcrusade/ Example virtual world http://secondlife.com/ Sociology research questions

  • How do networks within the ecosystem of groups

enable and constrain the formation of groups?

  • How do micro-group processes influence group

effectiveness and social identity?

Psychology research questions

  • What impact does playing video games has

people’s real lives?

  • Is online behavior different in MMOs vs.

tradition video games?

Macroeconomics research questions

  • lots of them

Computer Science research questions

  • quantitative metrics
  • algorithms
( ) ( )
  • i

b i

  • b

M M sim M M simila , ,

≠

∀ ≥

( ) ( )
  • ≤
  • −

− ∃

∈

thr t f t t f t

tw i i b
  • ffset
i
  • ffset
( ) ( )
  • i

b i

  • b

M M sim M M simila , ,

≠

∀ ≥

( ) ( )
  • ≤
  • −

− ∃

∈

thr t f t t f t

tw i i b
  • ffset
i
  • ffset
( ) ( )
  • i

b i

  • b

M M sim M M simila , ,

≠

∀ ≥

( ) ( )
  • ≤
  • −

− ∃

∈

thr t f t t f t

tw i i b
  • ffset
i
  • ffset
  • Fig. X. Group evolution
  • ver time

t 1 t 2 t 3

Impact on Social Science & Comp Sc

slide-42
SLIDE 42

12/2/2007 Jaideep Srivastava 42

MMORPG Example – Everquest 2

MMORPGs (MMO Role Playing Games) are the most popular of MMO Games

Examples: World of Warcraft by Blizzard and Everquest 2 by Sony Online Entertainment

Various logs of players’ behavior are maintained Player activity in the environment as well his/her chat is recorded at regular time instances, each such record carries a time stamp and a location ID Some of the logs capture different aspects of player behavior

Guild membership history (member of, kicked out of, joined, left) Achievements (Quests completed, experience gained) Items exchanged and sold/bought between players Economy (Items/properties possessed/sold/bought, banking activity, looting, items found/crafted) Faction membership (faction affiliation, record of actions affecting faction affiliation)

slide-43
SLIDE 43

12/2/2007 Jaideep Srivastava 43

DM Challenges for Social Science Research with Everquest 2 Data

Inferring player relationships and group memberships from game logs

Basic elements of the underlying social network such player-player and layer-group relationships need to be extracted from the game logs

Developing measures for studying player and group characteristics

Novel measures need to be developed that measure individual and group relationships for dynamic groups Novel metrics must also be developed for quantifying relationships between the groups themselves, the groups and the underlying social network as well as the groups and the environment

Efficient computational models for analyzing group behavior

Extend existing group analysis techniques from the social science domain to handle large datasets Develop novel group analysis techniques that account for the dynamic multiple group scenario as well as the data scale

slide-44
SLIDE 44

12/2/2007 Jaideep Srivastava 44

Measuring Player-Player Relationship (1)

Given multiple sources of player-player interaction such as traded with, teamed up with, chat data etc, how does one combine all this data into an accurate measure representing social interaction strength between players Naïve approach –

Reduce each interaction type into a single statistic, for example:

Traded with (TAB)- number of times A traded with B Teamed up with (QAB)– total time (in secs) A and B were teamed up together Chatted with (MAB)– total number of words exchanged between A and B

Represent social interaction strength between A and B by the vector SAB = [TAB QAB MAB] All future analysis work with the vector or a combination/aggregation

  • f the features
slide-45
SLIDE 45

12/2/2007 Jaideep Srivastava 45

Measuring Player-Player Relationship (2)

Key Issue with Naïve approach –

Different forms of interaction are indicative of different levels of social interaction strength which is unknown, i.e. it is not clear what is the relative weighting among the features of the vector SAB

For example if A and B exchange 100 messages and trade 20 times, it is not clear what is the difference in degrees of social interactions are associated with “100 messages exchanged” and “traded 20 times”

Traditional normalization techniques that account for magnitude difference do not solve this problem

Addressing the issue

Normalize the features of the vector SAB for an accurate measure of social interaction strength between players A and B It is desirable that the proposed approach is computationally efficient due to the size of the social networks involved in our domain Key Idea: The social interaction strength indications of the different features will be derived from the data itself

slide-46
SLIDE 46

12/2/2007 Jaideep Srivastava 46

Measuring Player-Player Relationship (3)

Approach 2

Let S* = {Sij | for all pairs of actors i and j} For each feature in the elements of S*, fit the corresponding values into two clusters C1 and C2 with means S1 and S2 respectively. It is expected that the cluster with lower mean (say C1) will consist of the weaker social interactions and the cluster with higher mean (say C2) will consist of stronger social interactions The midpoint between the boundaries of the two clusters will serve as a mid-point point for the range of social interaction strengths represented by that feature i.e. halfway between weak and strong social interaction strengths

slide-47
SLIDE 47

12/2/2007 Jaideep Srivastava 47

Measuring Player-Player Relationship (4)

Proposed Approach (Continued)

If Sm denotes the mid-point then all values of the feature in set S* can be transformed into a corresponding normalized measures

  • f social interaction strength using the transformation

f*={fij = sigmoid( d(fij – fm) )|for all pairs of actors i and j}

The above procedure is performed for each feature to obtain a normalized set of feature vectors The features now indicate equal levels of social interaction strength and standard vector based analysis or combination/aggregation techniques can be used safely

slide-48
SLIDE 48

12/2/2007 Jaideep Srivastava 48

Measuring Player-Player Relationship (5)

The main idea is that the clustering provides a rough sketch of the distribution of feature values across different social interaction strengths One of the inherent assumptions made is that the set S* covers the gamut of social interaction strengths in the underlying social network Not an unrealistic assumption if the set of actors is large enough If the set of actors is small then the social interaction strengths obtained will be relative to the set of actors considered –

In such a case the obtained social interaction strengths can still be used for certain important SNA techniques, such as centrality measures and community extraction, that are more reliant on the relative connection strengths rather than the absolute connection strengths

slide-49
SLIDE 49

12/2/2007 49

Trust in Social Networks

slide-50
SLIDE 50

12/2/2007 Jaideep Srivastava 50

What is Trust?

“Trust is a bet about the future contingent actions of

  • thers” (Sztompka 1999)

Main two components of Trust: Belief & Commitment

  • Alice trusts Bob if she commits to an action based on a belief that

Bob’s future actions will lead to a good outcome

Trust in not a single value: Context Dependent

  • If a trusts b then that does not mean that a has to always trust b.

Properties of Trust

  • Transitivity: Trust can be passed along a chain of trusting people.

Since Trust is not perfectly transitive, it could degrades along a chain of acquaintances

  • Composability: With information from many trusted people, there is

more reasoning and justification for the belief

  • Personalization and Asymmetry: Trust is a inherently a personal
  • pinion
slide-51
SLIDE 51

12/2/2007 Jaideep Srivastava 51

Key Terms

Trust propagation

An approach for inferring trust values in a network

Trust Metrics

An estimate of how much trust an agent a should accord an agent b, taking into account trust ratings from other persons in the network.

Trust in Online Social Network (Web of Trust)

Users either explicitly rate other people in their social network or trust values are inferred from them

slide-52
SLIDE 52

12/2/2007 Jaideep Srivastava 52

Trust Propagation

  • The Objective of trust propagation
  • A user trusts some of his friends, his/her friends trust their friends and so on…
  • Given trust and/or distrust values between a handful of pairs of users, predict

unknown trust/distrust values between any two users

  • When the source and sink are not connected,
  • look at the paths that connect them, use information from intermediate people to

make a trust recommendation

  • TidalTrust Algorithm (Golbeck, 2005)
  • Breadth First based search from source to sink
  • Search minimum possible depth from source to sink
  • Accept ratings from only the highest rated neighbours to sink
  • Use weighted average of trust (rating)
  • Issues in Trust propagation
  • Trust Decay: How much trust should be passed on to the next node.
  • Avoiding Cycles: Take the shortest path
  • Transitivity:
  • Partial Transitivity
  • Direct Trust vs. Recommendation Trust comes into play when discussing transitivity.
  • Normalization
  • “The normalized trust value from a person who has made many trust ratings will be lower

than if only one or two people had been rated.”

slide-53
SLIDE 53

12/2/2007 Jaideep Srivastava 53

Trust Metric

Network Perspective

  • Global: Compute a single value for trust from the global

perspective

  • Local: Compute values for trust from the perspective of individual

nodes.

Computational Locus

  • Distributed: Compute trust values based on neighbourhood
  • information. Distributed metrics are always global.
  • Centralized: A central 'authority' computes the trust values.
  • Privacy Issues: Assumes that trust values are publicly available.
  • Disadvantages of Distributed Approaches
  • Trust Data Storage
  • Convergence takes a long time
  • Advantages of Distributed Approaches
  • Trust values readily available.

Link Evaluation

  • Scalar: Compute trust between a and b.
  • Group: Compute trust between a set of individuals in V.
slide-54
SLIDE 54

12/2/2007 Jaideep Srivastava 54

Applications

Social Browsing Spam Filtering P2P Systems

Identifying the sources of 'good' data

Web of Trust

Semantic Web

Trust based Recommender Systems

slide-55
SLIDE 55

12/2/2007 Jaideep Srivastava 55

Application: Trust based Recommender Systems

Due to the propagation of trust over the social network, it is possible to compute the trust weight in more users and more items It reduces a cold start users problem It use trust information explicitly expressed by the users

  • the fake identities used for the attacks are not trusted explicitly by

the active users (and by the users she trusts)

slide-56
SLIDE 56

An Ant Colony Optimization Approach

Expert Identification in Social Networks

slide-57
SLIDE 57

Jaideep Srivastava

Expert Identification Problem

How does one route queries in knowledge markets which are

Set up like social networks One does not know the topic distribution or the expertise of the users beforehand The system is not centrally managing the system

Problem Statement

Given a graph of E experts and a topic distribution T, devise an approach for query routing that can be incrementally updated

slide-58
SLIDE 58

Jaideep Srivastava

Possible Solutions

Have a centralized repository of expertise and experts. Assumes that one already knows what the 'topics' are and who the corresponding experts are. Alternatively maintain a topic hierarchy over the network. Also assumes that the topics and that the topics are stationary. Desired qualities of a solution

identify experts routes queries without assuming a predefined topic scheme, allows for changing topics, expertise and topology, and does it efficiently

slide-59
SLIDE 59

Jaideep Srivastava

ACO Based Approach (1)

ACO (Ant Colony Optimization) Inspired from how ants forage for food

Initially an ant randomly forages for food When it finds a source it retraces its path Ants lay chemical trials called pheromones in their path which can evaporate if not reinforced Frequently used trails become stronger

Queries are represented as ants Whenever a query ant finds an answer to a query it retraces its path and lays out a trail Experts are the nodes with strong trails leading to them

slide-60
SLIDE 60

Jaideep Srivastava

ACO Based Approach (2)

Approach

For the first k iterations, queries are flooded in the network Queries are then be routed based on the scents Small amount of randomization is introduced in order to ensure that alternative routes (experts) are discovered Unlike traditional ACO techniques, different pheromones in our setting can be combined for cases where one encounters an unfamiliar query

Advantages

The 'solution' self-organizes 'Solution' can be incrementally built Graceful degradation of performance Can account for changes in the network Topics for expertise do not have to be predefined

slide-61
SLIDE 61

Jaideep Srivastava

ACO Based Approach

slide-62
SLIDE 62
slide-63
SLIDE 63

Jaideep Srivastava

A Patient Profile

  • Medical Problems
slide-64
SLIDE 64

Jaideep Srivastava

Background

In many cases people visit multiple doctors and specialists for their medical needs. The patients would be served better if there were better coordination between these specialists. To encourage cooperation one first has to identify clusters or groups of practitioners that people with certain characteristics are likely to go to Solution: find networks of referrals

slide-65
SLIDE 65

Jaideep Srivastava

Referral Networks and Cooperation

Problem:

Identify Referral Networks to encourage specialists to work together to offer better services.

Possible Solution

Offer incentives individually to specialists.

Defects in the Solution

Each specialist may want to “optimize” his/her own incentives. In such settings local optimization of services does not lead to global optimization of services

slide-66
SLIDE 66

Jaideep Srivastava

SNA Formulation of Poblem

Problem is reformulated as a SNA Problem Matrices can be constructed for variables of interest e.g., Patient- Specialist Matrix, Specialist-Location matrix etc. Techniques like Clustering, Link Analysis, Co-clustering, identifying cliques can be used.

slide-67
SLIDE 67

Some Emerging Applications

slide-68
SLIDE 68

12/2/2007 Jaideep Srivastava 68

Key Idea: Approach: Status and Future Work Key Benefits

Yahoo’s My Web: Me, My Interests and My People Yahoo’s My Web: Me, My Interests and My People

What does MyWeb represent?

  • What does creator think about a page?
  • What do I think about the page?
  • What do others think about the page?

What can be inferred?

  • Who are the community of people who are

“voted” as good resources on a topic?

  • What are the community of pages which are

“voted” as good resources on a topic?

  • Who are people/pages authoritative on a topic.
  • Improve Webpage ranking
  • Discovering communities of people and

Webpages based on what users think

  • Discovering expert Webpages and people on

given topics

  • Personalized Web and Community
  • Excellent source for personalized ads.

Current Ranking Schemes:

  • Creator Based Ranking.

Future Work:

  • Use of User Votes to improve ranking
  • Determining a most resourceful person.

{tags} {tags} {tags} {tags} {tags} {tags}

Community Aware PageRank PageRank Tag Aware PageRank

{tags} P1 P2 P3

slide-69
SLIDE 69

12/2/2007 Jaideep Srivastava 69

Key Idea:

  • 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 Approach: Status and Future Work

  • Develop a PageRank style scoring

scheme for ranking experts for various topics

  • Develop efficient algorithms for the same
  • Do we penalize users for possible ‘bad’

answers? If so how do we identify bad answers?

Key Benefits

  • The study of trends among questions

answers posted by the users esp. comparing behavior of the experts and non-experts

  • The above study as well as retaining

the experts can help improve the service provided by Yahoo! Answers

Yahoo! Answers: Identifying the Experts Yahoo! Answers: Identifying the Experts

Answer_1

User_1

Answer_2

User_2

Answer_n

User_n

User_x User_y User_z User Votes

Question

slide-70
SLIDE 70

12/2/2007 Jaideep Srivastava 70

Key Idea:

  • Current recommendation models

assume all users’ opinions to be independent, i.e. the i.i.d assumption

  • Can we make use of the social

network data of actors to relax this i.i.d assumption

Approach: Status (Research Issues)

  • Statistical Techniques exist for relaxing the i.i.d
  • assumption. Eg. Multilevel modeling and Random

mixed effects models

  • Research effort needs to be directed towards

extending or integrating the ideas presented in these techniques with existing recommendation systems

  • Alternatively, one can also work towards

designing complex graphical models for the proposed problem

Key Benefits

  • Understanding the impact of social

networks on market behavior

  • Improved recommendation systems

Influence of Social Networks on Product Recommendations 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

slide-71
SLIDE 71

12/2/2007 Jaideep Srivastava 71

trelease trelease

Approach: 1. Define feature vector, Mo, for objective movie

  • genre, MPAA rating, distributor, cast

2. Use feature vector as the basis to cluster movies 3. Take clustered movies as the training data to do classification for the new movie 4. Find the closet movie’s popularity function, fb where f is normalized 5. Get the current popularity function (query statistics) for the new movie

  • related queries include, e.g., movie’s name, stars

6. Use pattern matching to compute the distance between the objective movie (new one) and the similar movie (old one), and further to verify if the new movie is popular for each region in each time (interval) if not exists, increase ad. Example:

Using Query Statistics to Help Movie Advertisement Using Query Statistics to Help Movie Advertisement

( ) ( )

  • i

b i

  • b

M M similarity M M similarity , ,

≠

∀ ≥

( )

( )

  • ≤
  • −

− ∃

  • ∈

threshold t f t t f t

tw i i b

  • ffset

i

  • ffset

t t t t # queries # queries # queries # queries

MN CA Queries related to “Harry Potter” II trelease trelease I

as popular as usual in MN need more ad. in CA

slide-72
SLIDE 72

12/2/2007 Jaideep Srivastava 72

Summary

Research in Social Network Analysis has significant history

Social sciences: Sociology, Psychology, Anthropology, Epidemiology, … Physical and mathematical sciences: Physics, Mathematics, Statistics, …

Late 1990s: computer networks provided a mechanism to study social networks at a granular level

Computer scientists joined the fray

2000 onwards: Explosion in infrastructure, tools, and applications to enable social networking, and capture data about the interactions

Opens up exciting areas of data mining research

slide-73
SLIDE 73

12/2/2007 Jaideep Srivastava 73

Impact on Organizational Dynamics Research

Analyzing social networks

Hypothesis Testing and model verification

Example [Raymond2001]

  • Individual job performance was positively related to centrality in advice

networks and negatively related to centrality in hindrance networks composed of relationships tending to thwart task behaviors

  • Hindrance network density was significantly and negatively related to

group performance

Identifying who consults whom when confronted with a problem.

Such patterns can be identified by mining the chains of emails

Negative relationships may also be determined by mining changes in patterns of emails across time The building of trust among individuals can be studied over time

Analyzing knowledge networks

Mining patterns of knowledge in network These patterns may change/evolve over time

Social Networks data verification

Verifying subjective data (collected via surveys) with observed event sequences

slide-74
SLIDE 74

12/2/2007 Jaideep Srivastava 74

Impact on Organizational Policy Research

Data security

An absolute must

Privacy

Careful balance between privacy and data analysis

Impact of SNA on employee-organization relationship

Careful thought needed in managing this Should there be ‘opt-in’ or ‘opt-out’ options for employees? Is this too ‘big brother-ish’?

Bottom line

New technologies are radically transforming the workplace, impacting organization information flow like never before Not managed properly, they can lead to serious problems, e.g. employee releasing corporate secrets in blogs (Google) Need to have tools that enable the understanding (and thus management) of organizational information flow

slide-75
SLIDE 75

12/2/2007 75

Thank you! And be careful with that e- mail