GraphDB Group Meeting 2008-04-22

From Tetherless World Wiki

Contents 1 GraphDB Group Meeting 2 topic 3 Some Points 4 details

GraphDB Group Meeting

Part of: Jie Bao Blog
Date: April 22, 2008
Location: Lally 104
Participant(s): Jie Bao, Li Ding, Deborah L. McGuinness, Sanmay Das, among others

topic

Sanmay Das: Just a quick note about Tuesday's meeting: I'll be giving a (brief) talk about some work-in-progress with Malik on how Wikipedia pages become relatively stable and trusted sources of information. I'll review some existing work that proposes standard growth models for the accretion of edits to pages and explain how these models fall short in explaining Wikipedia. Instead of thinking about traditional growth models like preferential attachment we need to analyze Wikipedia pages as processes of _information growth_, which call for different models. I'll describe a couple of these models that we are working on and then spend some time presenting data from actual highly-edited Wikipedia pages to show how they converge to "stability."

Some Points

• Problem: how good content is on wikipedia?

• detect vandalism based on the text given for edits (logging annotation) -
  • good reason for extending to semantic logging!

details

Why Wikipedia work?
By sanmay das

1.Wilkison huberman 2007:
1.1 featured articles tend to have more edits
1.2 model for predict number of edits
2. Kittur et al 2007 chi
2.1 user groups also include vandalism and admin recovery
2.2 edits on psge in general go down
3. Stability model for most edited page
3.1 maximization the prob that new edit has new info
3.2 481 articles  more than 5000 edits
3.3 two types of edits: new content; vandalism
3.4 normalize data using alexa traffic statistics on wikipedia ( debatable)
4 result
4.1 credibility maximization process
4.2 edit decays   after giant pike
4.3 page visibility increase