Social Networks and the Semantic Web

1
Social Networks and the Semantic Web

• Peter Mika
• BI/FEW, BO/FSW
• Vrije Universiteit, Amsterdam

2
Contents

• Motivation I.
• The explosive mix of social networks and the Web
• A brief history of Network Science
• Social Network Analysis
• SNA in entrepreneurship
• Motivation II.
• Two problems in search of a solution
• Contributions

3
Motivation I.

• Excitement two vibrant application areas
  converging rapidly
• Social networks meet the Web
• The Semantic Web meets social networks

Hardly a surprise personal information made the
first Web
4
Social networks meet the Web

• Making Friendsters in High Places (Wired News,
  July 17, 2003)
• Will Microsoft Wallop Friendster? (Wired News,
  Nov 8, 2003)
• Social Nets Find Friends in VCs (Wired News, Nov
  17, 2003)
• Google spawns social networking service (CNET,
  Jan 22, 2004)
• The Technology of the Year Social Network
  Applications (Business 2.0, November 2003)
• And the backlash
• Social Nets Not Making Friends (Wired News, Jan
  28, 2004)
• Too many confirmations to answer
• Pretendsters, fakesters
• No clear goal
• Different definitions of friendship etc.
• Next generation social networks AND

5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
Links

• Friendship-network sites
• www.orkut.com
• www.friendster.com (over 5 million users)
• www.linkedin.com
• www.tribe.net
• www.tickle.com (formerly Emode)
• www.ryze.com
• www.flickr.com
• www.wiw.hu
• Weblogs
• Paolo Massa's blog
• Clay Shirky's blog
• Judith Meskill's blog

11
Friend-of-a-Friend

• Friend-of-a-Friend (FOAF) a standard vocabulary
  for recording personal information in a machine
  readable format (RDF)

• FOAF documents contain information such as
• name
• homepage
• image
• depiction
• interests
• projects
• publications
• memberships
• etc.

http//www.foaf-project.org/
12
Example

• ltfoafPerson rdfIDFrankvHgt
• ltfoafnamegt
• Frank van Harmelen
• lt/foafnamegt
• ltfoafmbox_sha1sumgt
• 241021fb0e6289f92815fc210f9e9137262c252e
• lt/foafmbox_sha1sumgt
• ltfoafhomepage rdfresource"http//www.cs.vu.nl/
  frankh" /gt
• ltfoafimg rdfresource"http//www.cs.vu.nl/fran
  kh/figs/FvH-2003.jpg" /gt
• ltfoafknows rdfresourcehttp//...HansAkkerman
  s /gt
• ltfoafknows rdfresourcehttp//...PeterMika
  /gt
• lt/foafPersongt

Thousands of these documents exist already on the
(Semantic) Web. All of them are linked together
through the knows relationship. This network is
also known as the FOAF-web.
13
Applications

• FOAF Explorer Text browser
• FOAF Naut graphical browser
• Codepiction browse the network in images

14
foafnaut
15
Codepiction
16
Codepiction
17
Network Science

• Graph theory meets real life
• Models to capture what is common of networks
  observed in the physical world
• Part of complex systems research in physics
• Broad applications
• Social Network Analysis
• Biology
• Chemistry
• Engineering

18
Graph theoretical concepts

• Degree of a node
• number of incoming / outgoing links
• Average Shortest Path, lav
• average shortest path over all pairs of vertices
  between which a path exists.
• Clustering Coefficient, C
• The clustering coefficient, C, represents the
  average fraction of existing connections between
  nearest neighbours of a vertex.
• Relative size of the largest component, S
• The relative size of the largest component is
  simply the size of the largest component divided
  by the total number of nodes.

19
Networks and graphs

• Euler (1736) bridges of Königsberg
• Later, Cauchy, Hamilton,
• Cayley,Kirchhoff, Pólya et al.
• Regular graphs
• Each node has the same degree
• High clustering, high paths lengths
• e.g. the lattice of atoms
• Nothing particular happened in the next 200 years

20
Erdos-Rényi

• Eight papers on Random graphs (1959)
• Given a fixed set of nodes.
• Choose two nodes and if you roll six with a dice,
  place a link between them.
• Poisson degree distribution (Not regular, but in
  large networks all nodes will have an average
  degree.)
• Low clustering, low paths lengths
• Explains important network phenomena
• six degrees of separation (next)
• the emergence of a giant component
• A tipping point when the average degree in a
  graph reaches 1 a giant cluster emerges
• Physics percolation or phase transition
• Dominates thinking over complex networks until
  recently

21
Poisson distribution
22
Power-law distribution

• Example Top 50 terms in TIME articles

• Also called Zipf-law in linguistics

23
Six degrees

• The (Stanley) Milgram experiment (1967)
• Two targets
• a stock broker in Boston, MA and the wife of a
  divinity grad student in Sharon, MA
• Starting points
• random people in Wichita, Kansas and Omaha,
  Nebraska
• Chain letter to forward to a personal
  acquaintance who is more likely to know the
  target person
• 42 of the 160 letters make it back, average chain
  is 5.5 long (overestimation!)
• six degrees of separation, small world
• Avg. distance increases logarithmically with
  network size
• Species in food webs (2), molecules in the cells
  (3), neurons in the brain of C. Elegans (14), The
  Web (19)

24
Granovetter

• Granovetter (1969) The strength of weak ties
• How people network (use their social
  connections) to find a job?
• Mostly through acquaintances (weak ties). Close
  friends are also friends of each other -gt unable
  to provide reach, access to new information
• A small and clustered world densely knit network
  of close friends acquaintances connecting them
• Impossible to explain by Erdos-Rényi (random
  graphs)
• high degree of clustering

25
Watts Strogatz

• Watts-Strogatz model (Nature, 1998)
• First successful attempt to reconcile clustering
  and random graphs
• Very few extra links cut back the separation
  drastically, while not alter the clustering
  coefficient significantly

26
Barabási

• The Barabási Model scale-free networks
• Growth Starting with a small number of nodes, at
  every time step, add a new node with m edges that
  link to m nodes already present in the system.
• Preferential attachment When choosing the nodes
  to which the new node connects, assume that the
  probability that a new node will be connected to
  a particular node depends on the degree of that
  node. Highly connected nodes are more likely to
  become even more connected. (The rich get
  richer, first movers advantage)
• The result is a network with a few highly
  connected nodes (hubs), while a considerable
  proportion of nodes have a degree of only one or
  two. In fact, the degree distribution ends up
  following a power law. (P(k) k-3)

27
Networks as competitive systems

• Barabási number of links is a function of time.
  First movers take all.
• How could new kids on the block succeed in the
  world of Barabási?
• Google, Boeing, Palm
• Barabási each node has a fitness.
• Preferential attachment is driven by the fitness
  connectivity product (fitness links)
• Speed of acquiring links is governed by fitness
• Still a power law?

28
Evolution

• Evolution of a network depends on the fitness
  distribution
• In fit-get-rich networks, power structure
  survives at all moments, the winners lead is
  never significant
• In some networks, the winner-takes-all. The
  fittest node gets almost all links, leading to a
  star topology. (1 hub, many tiny nodes)
• Example Microsoft ?

29
Criticism to the state-of-the-art

• Does this help us as individuals?
• Path finding is local search in real life
• We have 150 weak ties on average
• Would George Bush lend me his car?
• Not all ties are equal
• Effects of similarity, proximity lacking from
  models

30
Implications for architectures

• Network Robustness
• Error refers to random node failure.
• Attack refers to preferential removal of the
  most connected nodes.
• Robustness can be measured as the relative size,
  S, and the average shortest path, lav, of the
  largest cluster.
• Simulation shows that scale free networks are
  more robust to error but less robust to attack
  than both random and exponential networks.

31
Social Network Analysis

• Social Network Analysis (SNA) is the study of
  social relations among a set of actors. 1 2
• Network analysis is distinguished from other
  fields of sociology by a focus on relationships
  between actors rather than attributes of actors.
• sense of interdependence a molecular rather than
  atomistic view (network view)
• belief that structure affects substantive
  outcomes (emergent effects)
• SNA focuses on the various roles individuals play
  in social networks, the various kinds of
  relations that may exist and looks for
  viable/most efficient structures. SNA also scored
  successes in explaining network dynamics such as
  the spread of epidemics, fashions, inventions,
  technologies etc.

32
Network research in Entrepreneurship

• Application of SNA to entrepreneurial (business)
  networks
• i.e. nodes are entrepreneurs or their firms,
  links are business relations
• Applications to Business Venturing, Management,
  Strategy, Organizational Studies
• Key concept embeddedness
• Entrepreneurs are said to be embedded in their
  social network
• Business ties are said to be relationally
  embedded
• Focus on enterprise formation period or small to
  medium enterprises
• Investigates how certain properties of networks
  effect the success or failure of the enterprise
• Financial success
• Innovativeness

33
Research foci

• Network content
• Information and advice
• Emotional support
• Legitimation
• Relationship governance
• Trust
• Power and influence
• Threat of ostracism and loss of reputation
• Network structure
• Structural holes (Burt)
• Weak vs. Strong ties (Granovetter)
• Network as independent vs. network as dependent
  variable

34
Findings

• Embeddedness as a result of local search
• Lack of information, lack of resources to pursue
  broad searches
• previous linkages guide partner selection
• Social identity theory similarity strengthens
  self-image, similar people are treated more
  favorably
• Similarity can be cause (precondition) or result
  similarity breeds interaction vs. interaction
  breeds similarity

35
Two sides of embeddedness

• Enabling effect of strong ties
• Strong ties result in better information
  sharing/concerted acting, lower resistance
  (trustworthiness), comfort (familiarity),
  sometimes limiting of sharing (prevent leakage)
• Especially if first mover partners become tied
  up later
• Constraining effect of strong ties
• Over-embeddedness actors become locked-in,
  homogeneity (similarity) increases
• makes it more difficult to break out (discover
  information, resource niches) and diversify
  (learn, innovate)
• non-group members increasingly isolated
• No possibility for gaining advantage by brokerage
  (Burt)
• Contingency
• Different mix of strong vs. weak ties is
  appropriate
• at different stages of the enterprise
• for different activities

36
Motivation II.

• Two problems in search of a common solution
• How can the Semantic Web benefit from a machine
  understanding of the social networks of agents?
• In what way can Social Network science benefit
  from Semantic Web research and technology?

37
Problem cluster the SW perspective

• An ontology is a shared, formal conceptualization
  of a domain Gruber93.
• Ontologies build upon a shared understanding
  within a community.
• This understanding represents an agreement of
  experts over the concepts and relationships that
  are present in a domain.
• The social factor in ontology-based KM.
• Ontologies are expressed in machine processable,
  logic-based representations.
• This allows computers to manipulate ontologies.
• The machine factor in ontology-based KM.
• Yet, current SW technology ignores the social
  aspects of knowledge.
• Result hot potatoes, scalability problems

38
Hot potatoes

• Which one of these are considered difficult
  problems in ontology research?
• Ontology acquisition (automated)
• Ontology development (manual)
• Ontology evaluation and measures of ontology
  quality
• Ontology representation
• Ontology query
• Ontology-based platforms
• MAS, P2P, Grid, Web Services
• Ontology management
• Ontology storage
• Ontology-based reasoning
• Ontology versioning
• Ontology alignment, merging and mapping
• Ontology presentation and visualization
• Ontology-based search
• Ontology-based query
• Ontology-based collaboration
• And what do they have in common?

39
Scalability problems

• Ontologies for Information Management Balancing
  Formality, Stability, and Sharing Scope
  ElstAbecker2001

40
What does this mean for the Semantic Web?
41
Contingency

• An architecture of the Semantic Web that reflects
  the social nature of communication and knowledge

42
Problem clusterthe Social Science perspective

• A characteristic (and to some critics, a
  weakness) of research on networks is the lack of
  a core theory that in turn yields a set of
  well-defined propositions from which network
  constructs are defined
• The result is a loose federation of approaches
  (Burt, 1980) in which researchers often debate
  how concepts are operationalized rather than the
  underlying theoretical arguments themselves
  (Hoang and Antoncic, 2003)
• Result propositions are difficult to compare for
  their dependence on operationalization

43
Contribution

• A core ontology of social networks and
  relationships
• Formulate network theories in a semi-formal
  manner, explicate link between theory and case
  study data
• Case study the Semantic Web research community
• Action-oriented
• Beyond FOAF Towards a social structure for the
  Semantic Web

44
The End

• Your questions, my pleasure.

45
Visualizations

• Standalone tools
• Pajek
• http//vlado.fmf.uni-lj.si/pub/networks/pajek/
• NetMiner (commercial)
• http//www.netminer.com
• 3D
• Kinemages (3D data format)
• http//kinemage.biochem.duke.edu/
• several layout softwares
• Developers tools
• CAIDA (several tools)
• http//www.caida.org/tools/visualization/
• TouchGraph (Java API)
• http//www.touchgraph.com/index.html
• http//touchgraph.sourceforge.net/
• Jung (Java API for network analysis)
• http//jung.sourceforge.net/
• ATT GraphViz (layout sw, C/C)
• http//www.research.att.com/sw/tools/graphviz/