PeopleFinder: Searching for People, not just for Documents

1
PeopleFinder Searching for People, not just for
Documents
Technologies for Knowledge Sharing
ICT-Centre CSIRO Alistair McLean, Anne-Ma
rie Vercoustre, Mingfang Wu firstname.lastname
_at_csiro.au
Goal
To find out experts in a company, on a given
subject, and to be able to contact them.
To provide evidence that they are appropriate
experts.
Although documents are a valuable resource, for
many questions it is necessary to find the right
person rather than the right document.
Information Need
Search
Expert Details Evidence
A simple process of Web search
2
Example Searching for Experts in Virtual
Environments
Home page
More evidences
Corporate structure
Ranked experts for Virtual Environments
3
Benefits

• People Finder provides automatic and rapid
  identification of experts
  
• Without manual maintenance of employee
  expertise list
  
• Without any specialised user training
• With contact details and place in the
  corporate organisation chart
  
• With supporting evidence so the degree of
  expertise can be judged
  
• Decisions are based on the documents that staff
  load onto the corporate Intranet and projects
  they are involved in this automatically tracks
  expertise as staff develop.

Approach
People Finder leverages on structured and
unstructured information

• Unstructured information Intranet and
  extranet documents
  
• Structured information Corporate data
  (groups, projects description and memberships)

• Expertise evidence is based on
• Documents that contain people names
• Documents that are linked to project pages
  (not too far down)
  
• weight on Home pages, project pages and
  related pages (weight decreases with distance)

4
Evaluation
Preliminary Finding

• Test Collection (April 2003)
• CSIRO-CMIS Intranet/Extranet (html)
• List of staff, Home Pages
• Groups, Projects, Project descriptions (XML)
• Publications (XML)
• Business development contact database (mails)

• Leveraging on structure increases
  significantly the precision
  
• Adding more structured documents does not
  always increase precision

0.405
0.315
0.257
0.172
Base
0.209
0.187
0.159
0.125
Web
Web XML data
New
0.405
0.315
0.257
0.172
0.412
0.335
0.273
0.207
()
(93.8)
(68.4)
(61.6)
(37.6)
(1.7)
(6.3)
(6.2)
(20.3)
P
(3.2E-5)
(5.3E-6)
(4.6E-7)
(5.7E-5)
P
(0.84)
(0.31)
(0.23)
(0.0003)
Comparison between two collections (with and
without XML documents) by using the new system.
Average precision for base system without
structured data and new system over the Web
collection.