Online Social Networks Future Internet Design by tls14265

VIEWS: 6 PAGES: 242

									Online Social Networks & Future
Internet Design

S. Felix Wu
Computer Science Department
University of California, Davis

wu@cs.ucdavis.edu
http://www.cs.ucdavis.edu/~wu/



                                  1
     The value of the “Network”


• A concern about a network losing its “value”
  – while we are unsure about how to quantify the
    true value…




                                                    2
                Open Issues

• What is the “value” of this social network?
• How would this “value” be distributed and
  allocated to each individual peers?

• MySpace, Facebook, LinkedIn didn’t define
  the “game” for network formation and value
  allocation.
  – But, it is important to design the game such
    that the OSN will eventually converge to a
    state to best support the communities.



                                                   3
Social Network Games




                       4
             Fighter’s Club


• A Coalition game ~ like Warcraft
• Team members who are Facebook friends
  receive higher fighting powers

• ~1400 new friendships established daily
• ~10% of users with >95% friendships
  purely based on this game.



                                            5
        The Value of the OSN


• Networks ~ Applications
• Network value reduced ~ application value
  reduced

• How to operate the network to protect the
  value of the network?




                                              6
 Social Capital? Or, something else?

• The value might not simply determined by
  the properties of the network itself.
• “Perception matters”: how will the
  community members perceive the “image”
  of the network?

• Game theory analysis might or might not
  work well…
  – Network formation: value ~ topology



                                             7
 Facebook versus Personal Web Site


• WWW: everybody can see it
• Facebook: a selected set of people on your
  social network can see it




                                               8
9
10
            What is privacy?


• Tradition: specify who can’t see what!
  – Role-based Policy given by the domain
• Unsure: what is the positive/negative value
  of revealing something?
  – “Privacy” depends on many factors that are
    difficult to quantify.




                                                 11
    Community-based Networking


• Community is a set of networked nodes
  sharing some special “social” relationships.
• Value and privacy might be better defined
  within a community.
  – Consider the FB ecs30b group…




                                             12
           Urgent! Please contact me!
FROM:MR.CHEUNG PUI
Hang Seng Bank Ltd
Sai Wan Ho Branch
171 Shaukiwan Road
Hong Kong.

Please contact me on my personal box [puicheungcheungpui@yahoo.com]

Let me start by introducing myself. I am Mr. Cheung Pui, director of operations of
the Hang Seng Bank Ltd,Sai Wan Ho Branch. I have a obscured business
suggestion for you.

Before the U.S and Iraqi war our client Major Fadi Basem who was with the Iraqi
forces and also business man made a numbered fixed deposit for 18 calendar
months, with a value of Twenty Four millions Five Hundred Thousand United
State Dollars only in my branch. Upon maturity several notice was sent to him,…


                                                                            13
http://www.ebolamonkeyman.com/cheung.htm




                                       14
Pick your favor Spam Filter(s)




                                 15
            Antispam Filters

• An arm-racing game
  – In some sense, “the venders” are doing
    reasonable well.


• Bind the “spams” to the “sources”
  – IP addresses or Source Email addresses
  – Spammers control a large number of bots
  – Collaboration between yahoo.com and gmail.com




                                                16
This was considered a spam!




                              17
      This was considered a spam!




Sometimes, the cost of False Positive may be very high…
                                                      18
        The Implication of FP’s


• Spam-filters have to be conservative…
• We will have some false negatives in our
  own inboxes.

• We will use our own time to further filter..
  – For me, 1~2 seconds per email




                                             19
The emails I received just THIS morning…




                                           20
You have about 1 second to decide……




                                 21
22
23
24
            “Social Spams”


• They might not be spams as we often
  overlooked the social values of them!




                                          25
                Motivations


• What is the fundamental issue of “spams”?
  – Is it something to do with the design of our
    “basic communication mechanism”?


• Why can’t we explicitly utilize the “social
  context” in our communication?




                                                   26
            Davis Social Links


• What is the fundamental issue of “spams”?
  – Is it something to do with the design of our
    “basic communication mechanism”?
• Why can we explicitly utilize the “social
  context”?

• Routable identity versus receiver control
• Trust & Reputation system in “L3”


                                                   27
         Communicate: [A, D]



             B                           D



A                           C


As long as “A” knows “D’s routable identity”


                                               28
Hijackable Routable Identify




                               29
         [A,D] + social context



               B                                D



A                               C

“A” has to explicitly declare if there is any social
context under this communication activity with “D”!



                                                       30
    Internet & Routable Identity


• URL, IP address, email,….
  – For ANYWHERE in the Internet


• Without Routable Identity
  – Only available to certain parts of Social
    Networks


• Using OSN to perform access control


                                                31
    The same message content

• “M” from Cheung Pui

• “M” from Cheung Pui via IETF mailing list

• “M” from Cheung Pui via Karl Levitt




                                        32
             Social Context

• “M” from Cheung Pui
 Probably a spam
• “M” from Cheung Pui via IETF mailing list
 Probably not interesting
• “M” from Cheung Pui via Karl Levitt
 Better be more serious…



                                        33
             Social Context

• “M” from Cheung Pui
 Probably a spam
• “M” from Cheung Pui via IETF mailing list
 Probably not interesting
• “M” from Cheung Pui via Karl Levitt
 Better be more serious…
Either “M” is important, or
Karl’s machine has been subverted!

                                        34
     [A,D] + social context


           ??

            B                                  D



A                             C

    “A” has to explicitly declare if there is any
    social context under this communication activity
    with “D”! But, “D” only cares if it is from
    “C” or not!

                                                       35
        Online Social Network


• What is an online social network?
  – Realize and represent the human social
    networks “explicitly” (from “somewhat vague,
    fuzzy and implicit”)
  – Promote “OSN Applications”
  – Utilizing the “online” perspective to further
    develop the human social network


• Representation, Application, Development


                                                    36
37
38
39
Who is Salma?




                40
Who is Salma?




                41
Who is Salma?




                42
My message to Salma




                      43
The Social Path(s)




                     44
More Examples




                45
               CyrusDSL


• How do we accomplish these features?

• How do we realize the concepts scaleable?

• How will this system work against spams?




                                             46
         Just a couple issues …


• How to establish the social route?
  – How would “A” know about “D” (or “D’s
    identity”) ?


• How to maintain this “reputation network”?
  – MessageReaper: A Feed-back Trust Control
    System (Spear/Lang/Lu)




                                               47
 Social network analytical models


• Network Mathematics
  – Random graph model (low diameter)
     • Newman/Watts/Strogatz, 2002
  – Small world model (high cluster coefficient)
     • Watts/Strogatz, 1998
  – Scale-free network (node degree distribution)
     • Barabasi/Albert, 1999


• What is the right model for the network?


                                                   48
     [A,D] + social context


           ??

            B                                  D



A                             C

    “A” has to explicitly declare if there is any
    social context under this communication activity
    with “D”! But, “D” only cares if it is from
    “C” or not!

                                                       49
            Search on “OSN”


• How to get to       from         ?

• The Small world model
  – 6 degree separation (Milgram, 1967)
  – “existence of a short path”
  – How to find the short path? (Kleinberg, 2000)




                                                    50
           Social Network Analysis

1. Degree Centrality: The number of direct connections a node has.
     What really matters is where those connections lead to and how they
     connect the otherwise unconnected.


2. Betweenness Centrality: A node with high betweenness has
     great influence over what flows in the network indicating important links
     and single point of failure.


3. Closeness Centrality: The measure of closeness of a node which
     are close to everyone else. The pattern of the direct and indirect ties
     allows the nodes any other node in the network more quickly than anyone
     else. They have the shortest paths to all others.


4. Eigenvector Centrality: It assigns relative scores to all nodes in
     the network based on the principle that connections to high-scoring
     nodes contribute more to the score of the node in question than equal
     connections to low-scoring nodes.


                                                                             51
             Random Graphs

• G(n, p): n nodes and each edge with prob p




                                               52
            Random Graphs


• G(n, p): n nodes and each edge with prob p
• When p < 1/n, disconnected components
• When p is sufficiently large, 1 giant
  component
• How about diameter?
  – The maximum distance (in hops) between any
    two nodes.




                                                 53
    Random Graph (Erdos/Renyi)


• Probabilistically, each node has (N-1)p
  direct neighbors ~ Z
• ZD = N (D is the diameter)
• D = logN / logZ


• In two hops, each node will have Z2
  neighbors in (equal) probability?



                                            54
           Small World Model


• Low Diameter
  – Logarithmic or poly-logarithmic to N


• “High” Cluster Coefficient
  – cluster coefficient: the portion of X’s
    neighbors directly connecting to one of X’s
    other neighbors




                                                  55
            Cluster Coefficient


• Mesh network: Ccluster = 1
• Lattice Network (with degree K): Ccluster = 0
  – E.g., a linear line


• How about Ccluster for Random Graph?




                                             56
          Re-wiring (Watts/Strogatz)
Structured/Clustered
                       62




                            35



               Trade off between D and Ccluster !

                                                    57
        The model of Chung/Lu

• Weight wi for vertex i is the expected
  value of the number of edges connected to
  i
• The probability, for each pair of vertices i
  and j, to have an edge is
                           wiw j
            P(Lij  1)    n

                           w    k
                           k1




                                             58
              Open Issues

• Lots of analytical models, but which ones
  are better to model what?
• For instance, SN and OSN are quite
  different:
  – SN: expected D remains constant as graphs
    grow.
  – OSN: expected D might grow as the cost to
    new links might be almost zero.
• Real-world Data Validation is very much
  needed.



                                                59
  Two Issues about Low Diameters

• Why should there exist short chains of
  acquaintances linking together arbitrary
  pairs of strangers?

• Why should arbitrary pairs of strangers be
  able to find the short chains of
  acquaintances that link them together?




                                             60
        Routing in a Small World

•   Common question: do short paths exist?




• Algorithmic question: assuming short paths exist.
  How do people find them?
                                                  61
Kleinberg’s Basic setting




                            62
                    p, q, r


• p: lattice distance between one node and all
  its local neighbors
• q: number of long range contacts
• r: inverse probability [d(u,v)]-r
  – What is the intuition about r?
  – What about r = 0




                                            63
             Kleinberg’s results

A decentralized routing problem
  – For nodes s,t with known lattice
    coordinates, find a short path from s to t.
  – At any step, can only use local information,
  – Kleinberg suggests a simple greedy
    algorithm and analyzes it:




                                             64
           Local Information


• Local contacts
• Coordinate for the target
• The locations and long-range contacts of all
  nodes that have come in contact with the
  message.




                                            65
                    Results


• If r = 0, expected delivery time is at least
  a0n2/3.
   – Lower bound
• If r = 2, p = q = 1, a2(log n)2
   – Martel/Nguyen’s newer results
• 0 <= r < 2 ~ arn(2-r)/3
• r > 2 ~ arn(r-2)(r-1)



                                             66
                 Kleinberg’s Model

• Kleinberg’s model:
  – People  points on a two
    dimensional grid.
  – “P” Grid edges (short range).
  – “Q” long range contacts chosen
    with the inverse rth-power
    distribution.
  – How to search?
     • [S, T]
     • Find the neighbor closest to T


  – Work well only when r=2, p=q=1


                                        67
              Kleinberg’s Model

• Use only Local information, except the
  distance to the target.
  – However, what is the “global distance” in cyber
    space? Yet, the assumption behind is that the
    “edges” depend on the “relative distance”.




                                                      68
                        X, Y, and Z

• How will we tell whether the relative
  distance between X&Y is closer than X&Z?
  – X, Y, Z (assuming they are all direct friends to
    each other)
• One simple idea: “Keyword intersection”
  – KW(X), KW(Y), KW(Z)
  – 1/(#[KW(a) KW(b)] + 1)

  – Will this work? How about global distance?

      
                                                   69
Similarity




             70
         Kleinberg’s Lattice Model
• Graph embedded in a metric space (e.g., 2D lattice)
• “Search efficiently” using only Local information + long range
  contact(s)
   – ~ inverse probability [d(u,v)]-r
   – r = 2, a special case




                                                              71
                Some Extensions


• Hierarchical Network Models
• Group Structure Models
• Constant Number of Out-Links




“Small World Phenomena and the Dynamics of Information” by J.
  Kleinberg, NIPS, 2001




                                                                72
          Generation & Search


• There is a data structure behind and
  among all the social peers
  – Lattice, Tree, Group/Community
• The link probability depends on this “social
  data structure”
  – And, using it to generate the social network
• Searching may use “direct contacts” plus
  the knowledge about the social data
  structure


                                                   73
         Hierarchical Network Models

     • Representation
       – a complete b-ary tree, T
       – All social nodes are “leaves”
     • Distance and Link Probability
       – h(v,w) = the height of the least common
         ancestor of v and w in T
       – f (h(v,w)) probability proportional
     – f (h(v,w)) normalization in probability
          f (h(v, x))
         xv
     – k  c log 2 n out-degree in graph

                                                 74
     the Critical Value

            f (h)
      lim bh  0, 
       h 



            bh
                 

      lim f (h)  0, 
                       
     h 




                      h(v,w)
     f (h(v,w)) ~ b


                                   75
           Interpretation (1)

• /Science/Computer_Science/Algorithms

• /Arts/Music/Opera

• /Science/Computer_Science/Machine_Learning




                                         76
          Interpretation (2)


• Target: “stock broker @ Boston, MA”

• Next hop:
  – “bishop @ Cambridge, MA”
  – “banker @ New York City, NY”




                                        77
                Results

   1 (logn)
• Otherwise, no polylogarithmic search




                                         78
      How to Search in HNM??

h(v,w)
              h(v,w)
f (h(v,w)) ~ b
 f (h(v,w))
 f (h(v, x))
xv

k  c log n
          2




                               79
                    Useful Neighbor

                    Is “v” useful to reach “t”?
vt
v,t  T                                           T
commonAncestor(v,t)  u
                       ,root(T   u
Height(T   i,u  T 
          )                     )
                                  
Height(T   (i 1),t  T  t  T 
          )
                                       
                                             v        t

                                                          80
                    Useful Neighbor

                    Is “v” useful to reach “t”?
vt
v,t  T                                               T
commonAncestor(v,t)  u
Height(T   i,u  T 
          )            ,root(T   u
                                )
                                                  u
                                                      T 
                                  
Height(T   (i 1),t  T  t  T 
          )
                                       

                                           v               t

                                                                 81
                    Useful Neighbor

                    Is “v” useful to reach “t”?
vt
v,t  T                                               T
commonAncestor(v,t)  u
Height(T   i,u  T 
          )            ,root(T   u
                                )
                                                  u
                                                      T 
         
Height(T   (i 1),t  T  t  T 
          )                                          
                                                       T 
                                       

                                           v    w       t

                                                            82
                    Useful Neighbor

                    Is “v” useful to reach “t”?
vt
v,t  T                                               T
commonAncestor(v,t)  u
Height(T   i,u  T 
          )            ,root(T   u
                                )
                                                  u
                                                      T 
         
Height(T   (i 1),t  T  t  T 
          )                                          
                                                       T 
                                       

                                           v    w       t

                                                            83
          Useful Neighbor Recursively

                    Is “v” useful to reach “t”?
vt
v,t  T                                               T
commonAncestor(v,t)  u
Height(T   i,u  T 
          )            ,root(T   u
                                )
                                                  u
                                                      T 
         
Height(T   (i 1),t  T  t  T 
          )                                          
                                                       T 
                                       

                                           v    w       t

                                                            84
                 Search


• Find one “useful” neighbor in G as the next
  step

• What happens if NO useful neighbor?
• Expected steps to reach “t”.




                                            85
Probability to have 1 U.N.

                   log n

 Z   bh(v,x )   (b 1)b j1b j  log n
      xv            j1

                
 b i1leaves T 
  bi
              One leave
 log n
          bi       1
 b 
   i1
               
        log n blog n
         1 c log 2 n
 (1          )       n     All out-links
       blog n

                                               86
                  HNM


• High probability to be useful
• How about “constant links”?




                                  87
                    Group Structures


     • R is a group; R’ is a strict smaller subgroup

         q  R  2,v  R  (v  R R) (q  R  R q)

     • R1, R2,R3,… all contain v, then

       i,( Ri  q)  (v  Ri )        Ri  q
                                      i

     • q(v,w): minimum size of a group containing
       both v and w

                                                        88
How to Search in Group Structure??

    q(v,w)
                     
  f (q(v,w)) ~ q(v,w)
    f (q(v,w))
    f (q(v, x))
   xv

   k  c log n
            2




                                     89
                                   Idea


          • (v, t)
          • R is the minimum-sized group containing both v and t.
          • With property (1)



            q
          • Then:R    2,v  R  (v  R R) (q  R  R q)


                 R (t  R ( 2 R  R  R )
                              )

                         How to define “usefulness” of v?

                                                                  90
                           Usefulness of v


          • (v, t)
          • R is the minimum-sized group containing both v and t.
          • With property (1)



            q
          • Then:R    2,v  R  (v  R R) (q  R  R q)


                 R (t  R ( 2 R  R  R )
                              )

                        x,(l(v,x) 1)(x  R
                                              )
                                                                  91
Probability to have 1 U.N.

                   log n

 Z   bh(v,x )   (b 1)b j1b j  log n
      xv            j1

                
 b i1leaves T 
  bi
              One leave
 log n
          bi       1
 b 
   i1
               
        log n blog n
         1 c log 2 n
 (1          )       n     All out-links
       blog n

                                               92
     Probability to have 1 U.N.


                                 log n
               1
      Z              j 1 ( j1)   2 log  n
         xv
             q(v, x) j1
                2            c log 2 n
      (1                 )                n 
             2 log  n







                                                        93
                Results

   1 (logn)
• Otherwise, no polylogarithmic search




                                         94
       Fixed Number of Out-Links

 • Relax “t” to “a cluster of t”
            T                                           T

                                                 m L
                                                 r  Cluster
                              
                                                 n  mr
  v             t                       Cl                      Cl
                                    v                       t
                               
                    r: Resolution            w                       x
                                                                     95
             Question #1


• Why can’t we just                               T
  treat “Cluster” as
  “Super Node” and we
  go home (by applying
                                           m L
  the HNM results)?
                                           r  Cluster
                          
                                           n  mr
                                  Cl                      Cl
                              v                       t
                                     w                       x
                                                               96
             Not necessarily




    Cl                 Cl           Cl
p                  v            t
         q                  w            x




                                             97
            Probability



                             2 h(v,w )
     f (h(v,w)) ~ (h(v,w)  1) b
     Z  2r






                                           98
              Question #2


• For any out-link of v, what is the
  probability that the end point of the out-
  link is in the same cluster of v?




                                               99
     Answer


     (0  1)2 b0  1
     1 r r 1
            
      Z    2r 2








                         100
                  Results


• If the resolution is polylogarithmic, the
  the search is polylogarithmic if alpha = 1.




                                                101
              A “Similar” Process


                             T

Coloring the Links       u
                             T 
                                
                              T 
               

                   v   w       t

              
                                     102
   Random Graphs and Search/Routing

• Multiple Analytical models
   – Capture certain perspectives with social interpretation.
   – But, depending on WHAT we want to capture.
   – For example, simple social topology, weighted topology, content
     correlation
• Efficient Routing
   – Theoretically good following a given analytical model, especially the
     statistical graph generation.
   – Heuristics/Greedy not always clear.
   – Need real-world data validation




                                                                             103
                        X, Y, and Z

• How will we tell whether the relative
  distance between X&Y is closer than X&Z?
  – X, Y, Z (assuming they are all direct friends to
    each other)
• One simple idea: “Keyword intersection”
  – KW(X), KW(Y), KW(Z)
  – 1/(#[KW(a) KW(b)] + 1)

  – Will this work? How about global distance?

      
                                                  104
Similarity




             105
Similarity




             106
        The challenges of DSL


• The generator model is a lot more
  complicated.
• How to validate the model is going to be
  fairly human-oriented ~ we can only try our
  best to do “semantic” interpretation.
  – For instance, how to model the “profiling key
    words” analytically?




                                                    107
      Social Network Formation


• “Bit torrent”
  – Tic-for-Tac ~ incentive to form the download
    and upload networks over time


• Game-theoretical Analysis
  – “Understanding and re-discover” the incentive
    mechanisms under existing systems
  – “Engineering new Incentive mechanisms” to
    form the “right/best” networks we prefer!


                                                   108
            3-player coalition game
(N  {1,2,3},v)
                                          1
S  2  v(S)  1
S  1  v(S)  0                      2             3

Super-additive?               1           1
Efficient allocation?

The value must be given   2       3   2             3
by the interpretation
                              1           1


                          2       3   2             3
                                              109
             Coalition Game

 • Game
                             (N {1,2,...,n},v)
 • Characteristic Function
                               v(S) ~ S  N
 • Super-Additive
                
(S,T  N)(S T  )  v(S)  v(T)  v(S T)
 • Payoff Allocation
                       
                             x  (x i ) iN  N

                                               110
              Payoff Allocation


• Efficient
                      {x   N |  x i  v(N)}
                                  iN
• Individually Rational
                          xi  v({i}),i  N
• Imputation Set     I(N,v) 
              
 {x   N |  x i  v(N)  x i  v({i}), i  N}
            iN 

          
                                             111
Social Route Discovery for A2D


                 ??

                  B                                  D



A                                   C

    Let’s assume A doesn’t have D’s “routable identity”
    Or, “D” doesn’t have a global unique identity!
    Then, how can we do A2D?


                                                          112
                      Finding


                 ??

                 B                                 D



A                                  C

    A2D, while D is McDonald’s!
    D would like “customers” to find the right route.
    “idea: keyword propagation” e.g., “McDonald’s”


                                                        113
              Announcing



                B                          D


                                     K: “McDonald’s”
A                               C


    Hop-by-hop keyword propagation



                                                114
              Announcing



                B                                D
                         K: “McDonald’s”

                                           K: “McDonald’s”
A                                C


    Hop-by-hop keyword propagation



                                                      115
                Announcing



                   B                                D
                            K: “McDonald’s”


          K: “McDonald’s”                     K: “McDonald’s”
A                                   C


    Hop-by-hop keyword propagation



                                                         116
                Announcing



                   B                                D
                            K: “McDonald’s”


          K: “McDonald’s”                     K: “McDonald’s”
A                                   C


    Hop-by-hop keyword propagation
    And, I know I am doing FLOODING!!

                                                         117
                    Now Finding



Q: McDonald’s
                         B                                D
                                  K: “McDonald’s”


                K: “McDonald’s”                     K: “McDonald’s”
     A                                    C

         Search Keyword: “McDonald’s”
         A might know D’s keyword via two channels
         (1) Somebody else (2) From its friends
         Questions: does D need an identity? Scalable?

                                                               118
119
120
121
122
   Phishing/Hijacking is the default


Application Test
Q: McDonald’s
                         B                                D
                                  K: “McDonald’s”


                K: “McDonald’s”                     K: “McDonald’s”
     A                                    C

         Search Keyword: “McDonald’s”
         Questions: is this the right Felix Wu’s?



                                                               123
            Application Tests


• Example 1: credential-oriented
  – “PKI certificate” as the keyword
  – If you can sign or decrypt the message, you are
    the ONE!
• Example 2: service-oriented
  – Service/protocol/bandwidth support
• Example 3: offer-oriented
  – Please send me your coupons/promotions!



                                                 124
          “Routable Identity”


• Application identity =M=> Network identity
• Network identity =R=> Network identity
• Network identity =M=> Application identity




                                           125
         “App/Route Identity”


• Application identity =M=> Network identity
• Network identity =R=> Network identity
• Network identity =M=> Application identity

• Keywords =(MF-R)=> “Multiple Paths”
• Application identity selection
• Network route selection



                                           126
Hijackable Routable Identify




                               127
Application Test ~ “Layer 3”




                               128
                          Finding


Application Test
Q: McDonald’s
                         B                                D
                                  K: “McDonald’s”


                K: “McDonald’s”                     K: “McDonald’s”
     A                                    C

         Search Keyword: “McDonald’s”
         Questions: is this the right Felix Wu’s?
         How to avoid/control flooding??


                                                               129
  Scalability - Avoid the Flooding

• As it is, every keyword will need to be
  propagated to all the nodes/links (but the
  same keyword will be propagated through
  the same link once possibly with different
  policies).

• The issue: “who should receive my
  keywords?”



                                           130
      Community-Keyword Model


• A Social Peer, P, has 3 keyword sets:

  – Attributes (ATTR)
  – Original Keywords (OK)
  – Propagating Keywords (PK)




                                          131
       Community-Keyword Model

• Attributes (ATTR)
  – Keywords describing P (the social node)
  – Decided/configured by the owner of P
• Original Keywords (OK)
  – Keywords announced by P (the social node)
  – Decide/configured by the owner of P
  – Each keyword is associated with a propagation
    policy (decided by the owner of P)
• Propagating Keywords (PK)
  – From its own OK and other direct neighbors
  – Each keyword is associated with a propagation
    policy
                                                    132
       Community-Keyword Model

• Attributes (ATTR)
  – Keywords describing P (the social node)
  – Decided/configured by the owner of P
• Original Keywords (OK)
  – Keywords announced by P (the social node)
  – Decide/configured by the owner of P
  – Each keyword is associated with a propagation
    policy (decided by the owner of P)
• Propagating Keywords (PK)
  – From its own OK and other direct neighbors
  – Each keyword is associated with a propagation
    policy
                                                    133
       Community-Keyword Model

• Attributes (ATTR)
  – Keywords describing P (the social node)
  – Decided/configured by the owner of P
• Original Keywords (OK)
  – Keywords announced by P (the social node)
  – Decide/configured by the owner of P
  – Each keyword is associated with a propagation
    policy (decided by the owner of P)
• Propagating Keywords (PK)
  – From its own OK and other direct neighbors
  – Each keyword is associated with a propagation
    policy
                                                    134
               in Community of Davis


                ??

                 B                             D



A                                C


    Who should receive the keyword announcement for
    “McDonald’s”?


                                                      135
             as the Social Peer

• Attributes:
  – {McDonald’s Express, 640 W Covell Blvd, # D,
    Davis, (530) 756-8886, Davis Senior High
    School, Community Park, North Davis}




                                               136
             as the Social Peer

• Attributes:
  – {McDonald’s Express, 640 W Covell Blvd, # D,
    Davis, (530) 756-8886, Davis Senior High School,
    Community Park, North Davis}
• Original Keywords:
  – {McDonald, Davis, California, DHS, North Davis,
    Happy Meal, 50% off Tuesday, Lobster}
• Propagating Keywords:
  – {McDonald, Davis, California, DHS, North Davis,
    Happy Meal, 50% off Tuesday, Lobster, Anderson
    Plaza, Save-Mart, Taqueria Guadalajara}

                                                137
          “Per-Keyword Policy”


• For each keyword, we will associate it with
  a propagation policy: [T, N, A]
  – T: Trust Value Threshold
  – N: Hop counts left to propagate (-1 each step)
  – A: Community Attributes
• Examples:
  – [>0.66, 4, “Davis”] K via L1
         
  – [>=0, ,  ] K via L2



                                               138
               in Community of Davis


                ??

                 B                             D



A                                C


    Who should receive the keyword announcement for
    “McDonald’s”?


                                                      139
     Scalability & Controllability


• McDonald’s doesn’t want to flood the whole
  network
  – It only wants to multicast to the “Target set”
    of customers
• And, it only wants this target set of users
  being able to use that particular keyword
  to contact.
  – Receiver/owner controllability



                                                 140
        Autonomous Community


• Each social entity configures a set of
  “attributes” for itself.
• Some or all of the attributes will be
  exchange with certain neighbors.




                                           141
    Social/Community Attributes


                ??

                 B                             D



A                                C


    Who should receive the keyword announcement for
    “McDonald’s”? Answer:


                                                      142
      Relevant Attribute/OK/PK


• ATTR = Davis
• OK = McDonald’s
• PK = McDonald’s

• The owner uses the “policy” to control the
  flooding:
  – K = McDonald’s
  – [T > 0.66, N = 6, ATTR = “Davis”]


                                           143
             IP versus DSL

• IP address prefixes announced by BGP to
  ALL the Autonomous Systems in the whole
  Internet
  – Every IP node can send packets to McDonald’s at
    Davis (if we have a unique IP address)
• DSL will only announce “McDonald’s” (under
  the control of McDonald’s express) within
  the Davis social community
  – Only the receivers of the announcement can use
    the keyword to contact McDonald’s express!



                                               144
        Community-Keyword Model


• A Social Peer, P, has three keyword sets:

  – Attributes (ATTR)
  – Original Keywords (OK)
  – Propagating Keywords (PK)


• Flooding Avoidance + Receiver/Owner
  Control


                                              145
    [T >= 0, N =   , ATTR =  ] K
• What is the consequence?
  – Spam
                   
  – Denial of Service
• How to deal with it?



                                     146
      [T >= 0, N =      , ATTR =  ] K
• Limited Resources on PK
    – “P” can only remember up to M keywords in its
    
      own PK
•
                    between K
    Ordering Preference                    and Kj
                                       i



    – T(Ki) > T(Kj)
    – N(Ki) < N(Kj)
    – ATTR(Ki) ATTR(Kj)
                 
• Incentive Model
    – P is willing to pay a price



                                                    147
           Potential Problems


• Mostly only local contacts
  – Local interests dominate
  – Possible resource allocation for different
    ATTRs within the same community




                                                 148
              Community


• A connected graph of social nodes sharing
  a set of community attributes




                                          149
    Community


    ??

    B               D



A               C




                        150
Community Control:



                  D



  C                                              E
      Who should receive the keyword announcement for
      “wu@cs.ucdavis.edu”? Answer:

      Who should receive the keyword announcement fot
      “South Lake Tahoe Tournament”? Answer:
                                                        151
              Community


• A connected graph of social nodes sharing
  a set of community attributes




                                          152
    Community


    ??

    B               D



A               C




                        153
    Social/Community Attributes


                ??

                 B                             D



A                                C


    Who should receive the keyword announcement for
    “McDonald’s”? Answer:          but not ALL


                                                      154
                   Community


• A connected graph of social nodes sharing
  a set of community attributes
• The community members can decide the
  administrative policy within the community
  –   Membership maintenance
  –   Attribute setting
  –   Keyword propagation policy (e.g., allocation)
  –   Application-dependent policy
  –   Incentive model


                                                      155
            Potential Problems

• Mostly only local contacts
  – Local interests dominate
  – Possible resource allocation for different
    ATTRs within the same community

• “Reachability”
  – How likely will my keywords be able to go
    through to the community I want?
  – I must be a direct friend of the community
  – How can we set up “remote long range contact”?



                                                 156
       Community Development


• How will each one of us set up our
  Attributes and Original Keywords plus
  policy such that together we can
  communicate with each other “optimally”?
  – A game theoretical setting problem for network
    formation




                                               157
    Community


    ??

    B               D



A               C




                        158
    Network Formation


       ??

        B               D



A                 C




                            159
        Network Formation


              ??

               B                                D



A                               C




What is B’s incentive in adding the new ATTR keyword?

                                                    160
          Network Formation


                ??

                B                            D



A                                  C




    If B adds        , then A will add   !

                                                 161
           Network Formation


                  ??

                   B                                 D



  A                                  C



Both A & C: why would A & C be willing to establish a direct
friendship?

                                                          162
              Open Issues


• What is the “value” of this social network?
• How would this “value” be distributed and
  allocated to each individual peers?




                                           163
What is the “value” difference?

         B                        D



A                    C



         B                        D



A                    C

                                      164
    “C can join       !“

        B                  D



A                 C



        B                  D



A                 C

                               165
“A alone can help C to join more
         communities!“
          B                        D



A                     C



          B                        D



A                     C

                                       166
    Value Allocation for   B?
          B                     D



A                     C



          B                     D



A                     C

                                    167
    Nash Equilibrium with CS

           B                        D


        0~30~30
A                     C



       Propagating        or not?




                                        168
          Three Person Coalition Game

 (N,v, ),v  60u1,2  60u1,3  60u2,3 108 u1,2,3
   nf




Player 2 get “44”!                  1           1


                                2       3   2             3

Again, players 1 and 3 can          1           1
collaborate and break their
links with 2 to get “30” each
from merely “14”!               2       3   2             3

                                                    169
                Value calculation


    (S)v(S)  v(N)
S 2 N \{}

  (1,2)  60   (1,3)  48   (2,3)  30   (1,2,3)  72
  1  (1,2,3)
               (60  48  30)   (1,2,3)  72
        2
  1  (1,2,3)
               (138)   (1,2,3)  72
        2
 69  3   (1,2,3)
 72  v(N)


                                                            170
               Open Issues

• What is the “value” of this social network?
• How would this “value” be distributed and
  allocated to each individual peers?

• DSL, Facebook, LinkedIn didn’t define the
  “game” for network formation and value
  allocation.
  – But, it is important to design the game such
    that the OSN will eventually converge to a
    state to best support the communities.



                                                   171
Social Network Games




                       172
             Fighter’s Club


• A Coalition game ~ like Warcraft
• Team members who are Facebook friends
  receive higher fighting powers

• ~1400 new friendships established daily
• ~10% of users with >95% friendships
  purely based on this game.



                                            173
        The Value of the OSN


• Networks ~ Applications
• Network value reduced ~ application value
  reduced

• How to operate the network to protect the
  value of the network?




                                          174
      Let’s come back to SPAM!


• How will the proposed DSL model handle
  spam?
• Social Network games can be another
  major “social spams” to reduce the value of
  our online social network.




                                           175
      Let’s come back to SPAM!


• How will the proposed DSL model handle
  spam?




                                           176
       wu@cs.ucdavis.edu +


                 ??

                 B                                 D


                                             K: “wu@…” + Policy
A                                 C



    Who should receive the keyword announcement for
    “wu@cs.ucdavis.edu”? Answer:

                                                        177
            Even if “A” claims


                 ??

                 B                                D


                                              K: “wu@…”
A                                 C



    Who should receive the keyword announcement for
    “wu@cs.ucdavis.edu”? Answer:

                                                      178
                “B” can help…


                  ??

                   B                              D


                                               K: “wu@…”
A                                    C



    What is B’s incentive? What is B’s risk?



                                                      179
Message Value & Prioritization
      Link Ranks
       Reputation
       Incentives        Application IDS
   Other Trust Metrics




                           [good, bad] messages

                                                  180
181
            MessageReaper


• A Feedback Control Trust/Reputation
  system
• Trust needs to be maintained along the
  route path!




                                           182
                Reputation


• Adding “Trust” as another consideration in
  routing

• Per-packet Reputation Update
• Fast Stabilization
• Mobility without per-hop authenticated
  Global/Unique Network layer Identities



                                               183
       Reputation on Feed-back


               ??

               B                                   D



A                                 C


    “D” is the one to decide whether the message
    from A/B/C is good or bad!


                                                       184
           One Route path from A to D


Pkt[a>d]
A                B                    C               D


    End2End Trust:     “is this really from A?”

    RoutePath Trust:   “Should this path be used?”




                                                     185
           Basic Assumption about the Link


Pkt[a>d]
A                 B                 C                 D
    B & C have a way to decide whether they should
    establish a link between them, and they can
    authenticate each other:

            Secure MAC authentication
            Social Links in OSN
            Reputation-based Authentication
            Sybil Attack robustness

                                                     186
             The Attack Model
• Does the receiver really like this packet
  being delivered to me over a route path of
  links:
  –   Corrupted information
  –   Spam
  –   An incorrectly E2E-Authenticated packet
  –   Malware

• Assumption: the receiver has its own
  security policy to determine whether it like
  the packet/message or not!


                                                187
        D decides, and rewards/punishes…



                                             Pkt[c>d]
A                B                C              D


    Trust(B>A)       Trust(C>B)       Trust(D>C)
     Pkt[ab]         Pkt[abc]         Pkt[bcd]




                                                   188
                Trust Structure




We want to stabilize these decentralized values such that
they can be used to effectively choose the “best” route.


                                                            189
                  Game Theory Analysis



                                                        Pkt[c>d]
A                   B                     C                 D


    Trust(B>A)           Trust(C>B)            Trust(D>C)
     Pkt[ab]             Pkt[abc]              Pkt[bcd]



    Value Allocation: if a bad message is delivered, how
    should we distribute the “damage” along the route path?

                                                              190
      Trust Structure as the Utility




We want to stabilize these decentralized values such that
they can be used to effectively choose the “best” route.

                                                            191
 Three Trust Values per Relationship
                                u                    v

• Ta(u,v): u is directly connected to v. How
  much u trusts v?

• Ainit: v, as the initiator, sends a packet to u.
• Afwd: v forwards a packet to u . I.e., v is not
  the initiator of the packet.
• Art: sends a packet to, and, v forwards that
  packet to one of its other neighbors. And,
  the packet eventually reaches the
  destination.

                                               192
Example




          193
194
Routing with Trust




                     195
196
1000 nodes, 20% bad




                      197
1000 nodes, 10%/40% bad




                          198
Increasing the Spammers




                          199
Orkut (15329 nodes)




                      200
   Initialization versus Forwarding


• Do I believe that this message is being
  forwarded by one of my neighbors, and not
  being initialized from that neighbor?




                                         201
202
                Open Issues


• What is the “value” of this social network?
• How would this “value” be distributed and
  allocated to each individual peers?

• MySpace, Facebook, LinkedIn didn’t define
  the “game” for network formation and value
  allocation.
  – But, it is important to design the game such
    that the OSN will eventually converge to a
    state to best support the communities.


                                                   203
Social Network Games




                       204
     Using Trust as the “Utility”


• MessageReaper is being developed as a
  trust framework to protect the value of
  the network:
  – It’s not just about connectivity.
  – It’s about the quality of each individual
    relationship.




                                                205
    Collusive Attacks


       B                D



A                  C




                            206
    Robustness as OSN “Value”


           B                    D



A                     C

           B                    D



A                     C

                                    207
208
Davis Social Links over Facebook




                                   209
                   Smart Proxy


• Overlay Social Graph
• User-defined keywords
  and attributes
• DSL server
                             DSL
• Trust Routing Protocol




                           Facebook



                                      210
                Sub-communities


• Social Graph
• User-defined keywords
  and attributes
• DSL server
                             DSL
• Trust Routing Protocol




                           Facebook



                                      211
         Social Network Development


• Social Graph
• User-defined keywords
  and attributes
• DSL server
• Trust Routing Protocol     DSL




                           Facebook


                                      212
               “Bypassing” Facebook


• When you send a
  message…
   – Via Facebook
   – Via DSL
                             DSL
• Activity and Intensity
  hiding via
  Decentralization!

                           Facebook


                                      213
Community-Oriented Networking


• DSL offers a way to dynamically identify
  and establish social communities
  – But, we still have a lot of open issues


• Facebook:
  – Networks: email address dependent
  – Groups: you have to use your existing social
    network to invite.



                                                   214
215
               “Bypassing” Facebook
• When you send a
  message…
   – Via Facebook
   – Via DSL


• Activity and Intensity     DSL
  hiding via
  Decentralization!



                           Facebook


                                      216
DSL vs. Google




                 217
                 “Google”


• It’s about the “content”
  – Data-centric networking.
• Input to the Engine
  – A set of key words characterizing the target
    document.
• Output
  – A set of documents/links matching the
    keywords



                                                   218
                   “DSL”


• It’s also about the “content”
  – Application will decide the mechanism to
    further the communication.
• Input to the Decentralized Engine
  – A set of key words characterizing the target
    document (plus the aggregation keywords).
• Output
  – A set of DSL entities with the DSP (Davis
    Social Path pointer) matching the keywords


                                                   219
                 DSL Search Engine

Receiver or                                             Sender or
Content                                                 Reader




                     DSL Social World




    We are not just connecting the IP addresses!
    We are connecting all the contents that can be interpreted!

                                                                  220
              Google vs. DSL

• Google is essentially a “routing” framework
  between the contents and their potential
  consumers.
• Google decides how to extract the “key
  words” from your (the owner) web page or
  document.




                                            221
                 Google vs. DSL
• Google is essentially a “routing” framework
  between the contents and their potential
  consumers.
• Google decides how to extract the “key
  words” from your (the owner) web page or
  document.
• A DSL “owner/receiver to be” has the
  complete control over that. A balance
  between:
  – How I would like others to know about me?
     • And, I might want different folks to know me in
       different ways!
  – How I can differentiate myself from other Felix
    Wu?
                                                         222
             DSL is an old idea!
 A                                       B
We, as human, have been using similar
  communication principles. Maybe it is a
  good opportunity to re-think about our
  cyber communication system.
Identity is a per-application, context-
  oriented, and sometime relative issue.
Forming cyber communities of interests for
  application.

 A                F                          B
         F                       F


                                             223
LinkedIn: Get Introduced




                           224
Another one




              225
  DSL, Facebook, AL-BGP and GENI




 http://www.geni.net/DSLport

      AL-BGP over GENI/PlanetLab


Each DSL/FB user should
select a “closer” GENI
entrance as www.geni.net. In
other words, we might need to   Facebook
set up DNS records correctly.

                                           226
 DSL Architecture



Applications with Tests




         DSL




        AL-BGP
                          227
            Link



    Applications with Tests




        2          3
1
                              4


                                  228
AS-oriented Social Mapping



   Applications with Tests




                             229
    Control versus Data Path



      Applications with Tests




              control path

          2
1


                   data path
                                230
    Social-Control Routing



    Applications with Tests




          3

                   2
1



                              231
         DSL is still an old idea!
 A                                           B
Many applications already have “social
 network like” structure to enable P2P
 sharing across Internet.
 e.g., media sharing, on-line game,
 restaurant recommendation,…

Should we push these into a generic Social
 Network layer-3 to support all the
 applications?
  A                F                         B
         F                         F

                                             232
        A Different Internet?!
• Current Internet: every IP address will be
  able to communicate with every other IP
  address!
  – Allow by Default
• DSL-based “Internet”: we have a large
  number of “pairs” (two entities and their
  corresponding direct social link)
  – Deny by Default




                                               233
          The Physical Pipe


• Facebook, Overlay ~ no problem…
• Can we do better?




                                    234
                Comparison

• IP/email:
  – Convergence to an absolute consistent state
  – IP/email addresses are all you need, but the
    controllability is biased toward the sender
• DSL:
  – Convergence to a relative consistent state
  – No global network identity. Every DSL entity
    defines its own relative identity based on origin
    keywords.
  – Controllability is more balanced with other
    application challenges.


                                                   235
         Easy to Send & Receive
• Easy for both the good users and the
  spammers. (fair simplicity)
• The spammers abuse the “sending” right,
  while the good users have very limited
  options to counter back.
  – how easy can we change our email address?
  – how often do we need to do that?
• A “receiver” or “the owner of the
  identity” should have some control.
  – But, that means also “burden” to the users.

                                            236
         Easy to Send & Receive

• Easy for both the good users and the
  spammers. (fair simplicity)
• The spammers abuse the “sending” right,
  while the good users have very limited
  options to counter back.
  – how easy can we change our email address?
  – how often do we need to do that?
• A “receiver” or “the owner of the
  identity” should have some control.
  – But, that means also “burden” to the users.
                                            237
           Davis Social Links


• Peer-to-Peer System (P2P)
  – How human socially communicate?
• Online Social Network (OSN)
  – How to utilize OSN to enhance communication?
  – How to have a securer OSN?
• Autonomous Community (AC)
  – How to build/develop more effective
    community-based social networks?



                                              238
              Bike-MANET


•   The Wireless, Mobile Layer
•   The DSL Layer (keyword exchange)
•   Community Layer
•   Application Layer




                                       239
240
This research project examines fundamental issues in the current Internet architecture, namely "routable identity",
explicit trust/reputation control, and security/privacy considerations for online social networks such as Facebook and
SecondLife. It posits a dramatic change in Future Internet/networking design to facilitate future social communication
systems. The research builds and evaluates this new design, and examines the potential social impacts and insights
from it. The following research objectives are being studied: (1) Can a social peer generate, manipulate and protect all
layers of routability toward his/her own identity? (2) How should the notions of trust and reputation be explicitly and
formally represented and embedded in a large-scale communication architecture? (Trust/reputation become central in
a network architecture based on social identities). (3) What are the trade-offs between the utilization of the social
information by the network and the privacy protections needed for it?

For evaluation and validation, a social-network based communication system, named Davis Social Links (DSL), is
being built, over Facebook, both to mimic the human communication model and to integrate social trust relationships
into the network service infrastructure. Among its benefits, due to the routability design, the DSL architecture will
offer a new take on prevention of unwanted traffic (denial of service and so on), a costly and urgent problem in the
architecture as it is now.

The DSL project additionally investigates the communication model and its security and safety issues under virtual
reality social systems such as SecondLife. The ultimate goal of the DSL project is to study a dynamic, scalable, trust-
based, decentralized communication system/architecture for large-scale networks (10 million ~ 10 billion nodes).

This multi-disciplinary project is a joint effort among academic researchers and industrial collaborators from
computer science, sociology, statistics, and techno-cultural studies. The broader impacts of the research advance
potentially both the understanding and the innovation of communication networks. As communications networks exist
to support social purposes, the DSL research team carefully examines a number of implications regarding societal
relationships and how the Internet user community can directly benefit that relationship.




                                                                                                             241
        Acknowledgement
A                         B




A           F             B
    F                 F


                          242

								
To top