AnHai Doan

1
Managing Unstructured Data

• AnHai Doan
• University of Wisconsin-Madison

2
Unstructured Data ...

• Appears in many forms
• emails, Web pages, memos, call center text
  record, etc.
• Is pervasive
• 80 of the world data, and is growing
• Managed by many players
• SIGIR/WWW/KDD/AAAI, Google/Yahoo/Microsoft/IBM

We should work on it, or risk missing the
boat! But what sets us apart from the above guys?
3
Structure System Focus!

• Make it very easy to extract structures from raw
  data
• in raw form ? keyword search / bag analysis
• many apps want to go beyond that, they want
  structure
• we should encourage this ? back to our play
  ground
• not just DB IR, but DB IR IE
• Instead of working on isolated research problems,
  
  lets build end-to-end UDMS
• should repeat what we did with System R / Ingres
  system blueprint, followed by 20 years of rapid
  progress
• unifies accelerate our research efforts
• keeps work grounded, make impact

4
What Does this System Look Like?
Joe Hellerstein
Flexible modes of interaction
Extraction Integration
Joe Six-Pack
Mass collaboration Best-effort, pay-as-you-go,
improving over time Scale up to huge data (by
running over clusters)

• DB IR IE II, in a best-effort, Web 2.0
  fashion

5
Broader Impacts

• Great for many current applications
• e-science, business, personal data, Web data,
  etc.
• Great for many current research topics
• IR, integration, PIM, data spaces
• user interfaces, HCI, mashup
• provenance, uncertainty
• cluster management
• query processing
• monitoring, handling changes, pub/sub systems
• Raises novel research issues
• mass collab, best-effort, extraction, helping Joe
  Six-Pax
• Helps define data mgt principles in broader
  contexts