POLYPHONET: An Advanced Social Network Extraction System from the Web

1
POLYPHONET An Advanced Social NetworkExtraction
System from the Web

• Yutaka Matsuo Junichiro Mori
  Masahiro Hamasaki
• National Institute of Advanced University
  of Tokyo National Institute of Advanced
• Industrial Science and Hongo 7-3-1, Tokyo
  113-8656 Industrial Science and
• Technology Japan
  Technology
• y.matsuo_at_aist.go.jp
  jmori_at_mi.ci.i.u-tokyo.ac.jp
  hamasaki_at_ni.aist.go.jp
• (WWW2006)
• Finding Social Network for Trust Calculation
  (ECAI 2004)
• Yutaka Matsuo, Hironori Tomobe, Koiti Hasida and
  Mitsuru Ishizuka

2
ABSTRACT

• Social networks in Semantic Web
• Knowledge management,
• Information retrieval,
• Ubiquitous computing..
• POLYPHONET
• Extract relations of persons
• Detect groups of persons
• Obtain keywords for a person.

3
Introduction and Related work 1/3

• Social Network
• Please indicate which persons you would regard
  as your friend.
• Social networking services (SNSs)
• Friendster http//www.friendster.com/
• Orkut http//www.orkut.com/
• Imeem http//www.imeem.com/
• 3600 http//360.yahoo.com/
• Web of trust
• Ontology construction

4
Introduction and Related work 2/3

• Referral Web (1995)
• social network extraction system from the Web
• Two person X and Y by putting a query X and Y
  to a search engine.
• Flink
• online social networks for a Semantic Web
  community
• Given a set of names as input, the component uses
  a search engine to obtain hit counts

5
Introduction and Related work 3/3

• Name disambiguation probability model
• Co-occurrence information
• provided by a search engine
• to detect the proof of relations
• Google-Hacks book
• PageRank, HITS
• Web graphs
• Link structure of Web pages is seen as a social
  network.

6
Social Network Extraction 1/4

• Nodes and Edges
• Nodes a list of persons is given beforehand
• JSAI2003,JSAI2004,JSAI2005 and UbiComp2005
• Edges between of nodes are added using a search
  engine.
• Co-occurrence
• matching coefficient, nXY
• mutual information, log(nXY /nXnY )
• Dice coefficient, (2nXY )/(nX nY )
• Jaccard coefficient,(nXY /nXvY )
• overlap coefficient, (nXY / min(nX, nY )) ECAI
  2004
• cosine, (nXY / )

7
Social Network Extraction 2/4
K30
8
Social Network Extraction 3/4
9
Social Network Extraction 4/4
_
10
Advanced Extraction

• Relationship
• Relationships between people
• 30 kinds of relationships
• http//vocab.org/relationship
• POLYPHONET
• Co-author co-authors of a technical paper
• Lab members of the same laboratory or research
  institute
• Proj members of the same project or committee
• Conf participants in the same conference or
  workshop

11
Advanced Extraction - Class of Relation 1/2

• GoogleTop(X Y,5)
• C4.5
• Five-fold cross validation (JSAI Case)

High tf-idf terms manually categorize data set.
12
Advanced Extraction - Class of Relation 2/2
13
Advanced Extraction Scalability 1/3

• For example -
• The network density of the JSAI2003 social
  network is 0.0196 with o.2 threshold.

14
Advanced Extraction Scalability 2/3
_
15
Advanced Extraction Scalability 3/3
16
Advanced Extraction Intellectual link 1/6

• Intellectual link
• A relation between a pair of persons with similar
  interests or citations
• Evaluation
• They plot the probability that the two persons
  will attend the same session at a JSAI
  conference.
• Idea
• If two persons are researchers of very similar
  topics, the distribution of word co-occurrences
  will be similar.

17
Advanced Extraction Intellectual link 2/6

• Keyword extraction

Termex 37
18
Advanced Extraction Intellectual link 3/6

• Keyword extraction
• 567 researchers with 3981 pages
• They gave questionnaires to 10 researchers and
  defined the correct set of keywords.

19
Advanced Extraction Intellectual link 4/6

• X2
• idf
• hit

20
Advanced Extraction Intellectual link 5/6
21
Advanced Extraction Intellectual link 6/6
22
POLYPHONET 1/4
23
POLYPHONET 2/4
24
POLYPHONET 3/4
25
POLYPHONET 4/4
26
Conclusion

• This paper describes a social network mining
  approach using the Web and organize those methods
  into small pseudocodes.
• New aspects of social networks are investigated
  classes of relations, scalability, and a
  person-word matrix.
• This paper implemented every algorithm on
  POLYPHONET.

27
(No Transcript)
28
(No Transcript)