Value filtering

1
Value filtering

• Problem
• Vast amount of data
• Traditional method is not enough
• Hard to find the valuable information

2
Goal

• Help users with information search
• Attach value to documents
• Filter out garbage information

3
Outline

• Filtering model
• User-Filter interaction model
• Information domain
• Movie, Music, Web, Legal Cite, ...
• Filtering criteria
• Content-based, Context-based, ...
• Metadata architecture
• Value distribution

4
Filtering model

• Query/Response model

Filter
Query
Information
Response
5
Filtering model

• Push model

Filter
Useful Information
6
Filtering model
This way, Please!

• Guide model

?
7
Information domain

• Different domain -gt different method
• Legal cite, Entertainment, Newsgroup, ...
• Utility/cost
• ex) Legal cite Defense System Movie
• Technological barrier
• ex) Text document Movie

8
Filtering criteria

• Content-based
• Access history-based
• Context-based
• User-preference based

9
Content-based

• Similarity of document to users interest
• Keyword vector, LSI (Latent Semantic Index)
• Most systems
• Search Engines
• Tapestry (Xerox Palo Alto), FAB (Stanford),
  WebWatcher (CMU),

10
Access history-based

• Analysis of users access pattern
• Personal access pattern
• Personalized access history of each user
• Letizia (MIT Media Lab)
• Global access pattern
• Global access pattern to each document
• WebWatcher (CMU), Path-profile (Microsoft), KSS
  (Stanford), ...

11
Context-based

• Hyper-link Context
• Mostly for Guide Model
• KSS (Stanford), WebGlimpse (U of Arizona),
  WebWatcher (CMU)
• Social-Network Context
• Referral Web (ATT Lab)

12
User preference-based

• Personal profile matching
• Find similar taste user
• Global preference measure
• Find globally famous/popular document
• Explicit Feedback
• Users explicit feedback on the document
• Implicit Feedback
• Automatic extraction of users preference

13
User preference-based

• GroupLens (U of Minnesota), Ringo (MIT)
• Explicit user-profile matching (user voting)
• Google (Stanford)
• Implicit Global preference (Hyperlink)
• PHOAKS (ATT Lab)
• Semi implicit Global preference (Netnews)

14
Metadata architecture

• Architecture for Value exchange
• PICS (Platform for internet content selection)
• W3C Standardization effort
• Inspired by Adult Site Filtering
• Publisher, Rater, Filter
• User selects Rater
• Stanford ComMentor

15
PICS

• (PICS-1.1 http//ra/v1.0/
• labels
• on "1994.11.05"
• until 1995.12.31"
• for "http//c.s/g.html"
• by "John Doe"
• ratings (violence 3 sex 2))