A Platform for Personal Information Integration Xin Luna Dong, Alon Halevy lunadong, aloncs'washingt

1
A Platform for Personal Information
IntegrationXin (Luna) Dong, Alon
Halevylunadong, alon_at_cs.washington.uga.eduUniv
ersity of Washington

• Paper Review
• by
• Delroy Cameron
• 2/23/2006

2
A Platform for Personal Information Integration
Introduction

• Digital Information Explosion
• Search
• WWW Search (e.g. Google, Yahoo)
• Personal Search (Desktop data)
• SEMEX (SEMantic EXplorer)
• Goals of Personal Information Management (PIM)
• Uses Semantic Associations
• Browsing and Querying
• Automatic Creation/Detection of Association

3
A Platform for Personal Information Integration
1. Browsing by Semantic Association

• SEMEX Technique
• Requires Logical View instead of hierarchical
• Objects and relations between those Objects
• e.g. Person, Book, AuthoredBy, AttachedTo
• Instantiation of Logical View
• Association Database or Personal Information
  space
• Current Problem
• Keyword Based Search
• Data stored in Directory Hierarchies on PCs
• Time to traverse trees to find relationships

4
A Platform for Personal Information Integration
SEMEX Data Association Techniques

• Association by File type
• Simple Case
• Email Clients
• e.g. senders and recipients
• More Complex Case
• e.g. AuthorOf
• Associations by LaTex Types and PPT
• Association using External Source
• e.g. List of all Graduate Students in a
  University
• Association by Integration
• Multiple sources with simpler associations
• Spread Sheets, WWW
• Reconcile References

5
A Platform for Personal Information Integration
2. Automatic Creation/Detection of Associations

• Data Integration using SEMEX
• Import data into the users personal information
  space
• From www
• From local files
• Formatting Data
• Scraping from Files or Web pages
• Form Associations
• External sources and users personal Domain
  Model
• Import Data
• Reconcile references
• Analyze data for pattern matching
• Derive new associations

6
A Platform for Personal Information Integration
3. PIM Challenges

• Handling Long-lived/Evolving Data
• Data consistency, seamless updating
• Reference Reconciliation
• Schema Mapping
• Right Granularity for personal Data
• Keep models simple, users not technically savvy
• Develop user-oriented v. Database Design
• Let system fit user habitat
• Not fit user activities into Database
  environment
• Combining structured/structured data
• Seamless conversion oblivious to user

7
A Platform for Personal Information Integration
4. SEMEX Architecture

• Domain Model
• Ontology of personal information, with objects
  and their associations
• Data Repository
• Association database or Personal Information
  Space
• Reference Reconciliation
• Ontology of personal information
• Associations and Instances
• Simple already stored
• Extracted rich objects e.g. power point
• External Sources
• Defined similar to views in a database

8
(No Transcript)
9
SEMEX Architecture
10
SEMEX Interface
11
A Platform for Personal Information Integration
4.1 Browsing and Querying

• SEMEX Keyword Search
• All documents mentioning the keyword
• Returns Heterogeneous data,
• From many different Classes
• SEMEX Selection Queries
• Specified Class, and specify given Attributes
• SEMEX Association Queries
• Conjunctive over triplets, pair of objects and
  their relation
• Returns links much like web browsing
• e.g. Search all Bernstein publications

12
A Platform for Personal Information Integration
5. Reference Reconciliation

• Mesh External Data to users Domain Model
• e.g. Mike Carey, M. Carey refer to the same
  person
• Previous Techniques
• Reconciling tuple references in DB Table
• Assume References have same attributes
• Each attribute has a single value
• Challenges
• Heterogeneous Data, different set of attributes
• References have many attributes
• Each attribute may have multiple values

13
A Platform for Personal Information Integration
Reference Reconciliation contd

• SEMEX Approach
• Important Definitions
• Reference
• One or several representations of an object
• Class
• May have several Keys
• A Key is a set of Attributes that uniquely
  define an object in a class
• e.g Person Class email, fname lname Keys

14
A Platform for Personal Information Integration
Reference Reconciliation contd

• SEMEX Approach
• Pair wise Decisions
• Enrich references when they match with others
• Contain more information about the domain object
  
• Support more sophisticated decision matching
• An Enriched reference contains a set of values
• e.g. multiple spellings for last name
• References may be grouped
• e.g. Publications with multiple references to
  each of its authors
• Group so that all reference points to single
  author

15
A Platform for Personal Information Integration
Reference Reconciliation contd

• Algorithm
• Step1 Based on Shared Keys
• Merge input references on a key value
• e.g. Person email, fname, lname,
• Step2 - Based on String similarity
• Use edit distance to measure similarity
• An Independent heuristic, uses known format
• e.g. phone, email
• Step3 Applying Global Knowledge
• Time Series Comparison
• collects references judges to be similar, gets
  the time stamp and merges if there is little or
  no overlap
• Step4 - Search Engine Analysis
• Feeds text into Google and compares top hits

16
A Platform for Personal Information Integration
Objects from Multiple Classes
17
Case Study
18
A Platform for Personal Information Integration

• Preliminary Experiments

19
A Platform for Personal Information Integration
Reference Reconciliation contd

• Objects in Multiple Classes
• Reconcile each class in isolation
• Create a Dependency Graph
• A Node for each candidate pair of references
• Each node has similarity score (0 to 1)
• An edge between nodes mean we must reconsider
  similarity if we re-compute
• Using the Dependency Graph
• If c1 and c2 merged, pa1 and pa2 also merge
• pe1 , pe2, pe3, and pe4 may merge
• i1 and i2 merge

20
A Platform for Personal Information Integration
Reference Reconciliation contd

• Evolving Objects
• Publication may change author, title, etc
• Blurs the line of when to model as single or
  multiple reference
• Distinguish granularity
• Coarse Grain compile from fine-grained based
  on certain similarities

21
A Platform for Personal Information Integration
Automatic Integration

• Integrate External Sources
• Approach
• Import instances and associations from external
  source into personal information space
• Mark imported data as Temporary or Permanent
• Pose query to find intersection of data
• Export this intersected data to the spreadsheet
• Previous Work
• Schema Mapping

22
Browsing Associations with Semex
LUNA DONG
Different references to the same person
AuthorOfArticles
MentionedIn
SenderOfEmails
RecipientOfEmails
Reference reconciliation identifies all
references to the same real-world object
Coauthors
23
Who are Working on Semex? Keyword Search
Returns Associated Instances
Search Semex
3 Conferences for publishing Semex papers
105 Images in Semex papers
2398 Messages 2 Presentations 65 Articles
15 Persons working on Semex (though they are not
named Semex )
24
How do I Get to Know this Person? Semex
Provides Lineage Information
Susan Dumais
Latest Lineage
Shortest Lineage
User Do I know this paper of Susan Dumais?
Semex Yes, you once cited it.
The last time we mentioned Susan Dumais is in an
email
Earliest Lineage
I got to know Susan Dumais by citing her paper
25
A Platform for Personal Information Integration

• Thank You!