Context-driven Access to
Personalized Digital Multimedia Libraries
Invited Talk at the
1st International Conference on Digital Libraries New Dehli, India 24-27 February, 2004
Erich J. Neuhold
Fraunhofer IPSI
Darmstadt, Germany
Content
Context of Information Access
Personalization in Digital Libraries
Classification of Personalization Methods
Recommender Systems
Next Generation Personalization
Personal Web Context
Personal Reference Library
Cooperative Annotation
Context Definition
Context is:
The sum total of meanings (associations, ideas,
assumptions, preconceptions, etc.) that:
(a) are intimately related to a thing, (b) provide the origins for, and
(c) influence our attitudes, perspectives,
Context of Information Access
Tasks
Skills
Interests
context of the user
context of
information object
Links Annotations Metadata
Relationships
Environment
Ways of using Context
annotation create context for information objects
and can be used to support cooperation and improved retrieval
putting relevant information object into a working
context (structuring metadata)
personalization based on modeling the
Motivation: Personalization
“Digital libraries that
are not personalized
for
individuals will be seen as defaulting on their
obligation to offer the best service possible”
DL: Content-to-Community
Mediation
Understanding of users in domain
Providing
enrichment Facilitating
Challenges – (1): Targeting
Content
Content:
is becoming more voluminous is becoming more varied
Contributes to information overload
Community:
Is made of diverse individuals Conflict:
Individual-specific information need
Holistically targeting entire community
Challenges – (2): Understanding
Users
Users have a Context:
Differing cognitive patterns (i.e. skills,
interests)
Embedded in a Community Multiple tasks or goals
User have competing simultaneous roles that are:
Interactive
Related to other entities in a given
domain
Autonomous
Require individual conceptualization
of the information space
Personalization as a solution?
Interaction Autonomy DL user
Personalization as a Solution
Personalization dynamically adapts a system’s service or content offer, based on a model of the user, in order to better meet or support the preferences and goals of individuals and specific target groups [Riecken 2000]
Objective of Personalization:
Goal oriented information supply
Checkpoints: Meeting the
Challenges
User Models - complex, i.e. context-based:
Differing cognitive patterns (i.e. interests) Relations to other domain entities
Individual conceptualization Multiple tasks or purposes
Group Interaction:
Infrastructure supporting group interaction &
information needs
Personalization Methods and Cognitive
Patterns
There are several ways personalization can support user’s cognitive patterns:
Personalization DL
Services Content
Special
Service PropertiesService Enrichment
Selection Structuring
• Notification
• Personal Agents
• Configuration
• Visualization
• Recommendation
• Annotation • Information Filtering
• Container • Bookmarks • Navigation
Selection: Information Filtering
Information Filtering:
Selectivity from dynamic
information sources on behalf of a user
Dynamic Information Filtering:
Information Filtering in the
presence of rapidly
changing user interests
user informs the system of
their new interests
Special Services: Notification
ChangeDetect supports:
Email notification:
sends an automatic
email whenever pages are updated
saves your favorite web
pages
monitors content
FREE service
http://
www.changedetect.com
Enrichment: Recommendation
There are several types of Recommender Systems:
Collaborative Content-Based
Demographic based Utility-Based
Types of Recommender Systems -
Collaborative
Collaborative = user-to-user
Based on similar users’
ratings
Rate movies you have seen
Receive online movie
Types of Recommender Systems –
Content based
Content-Based = item-to-item
correlation between the item’s
content and user preference
Adaptive interface
Types of Recommender Systems –
Demographic and Utility based
Demographic-based:
First: categorize the user
based on personal attributes
Second: filter based on
similar demographic categories [Burke 2002]
Utility-based:
Computation made on
the utility of each item
for the user [Burke 2002] Filtered Items
Utility-based Filter
Resource Features
Demographic-based Filter
Category_01
match
f (user) 1 2 3 4 5 Grouped Resources
Information Seeker
Types of Recommender Systems –
Knowledge based
Knowledge-based:
uses functional
knowledge about how a particular item meets a user need [Burke 2002]
i.e.
Type of cuisine Price range
Types of Recommender Systems -
Hybrid
Hybrid Approach:
Combination of filtering
methods - current trend
Overcomes weakness
single methods
Improves system
performance Example:
Graph-based
Recommender
System for DL [Huang 2002]
Content-based
and
Demographic-based
Content-Based Filtering:
Correlation between similar books
Demographic-Based Filtering:
Checkpoint: Meeting the
Challenges
User Models - complex, i.e. context-based
considering:
Differing cognitive patterns (i.e. interests) Relations to other domain entities
Individual conceptualization Multiple tasks or purposes
Group Interaction:
Infrastructure supporting group interaction &
information needs
The Personal Web Context
It is not uncommon for people within a
community to discover resources (i.e. other persons, documents) via serendipitous means
because they are (directly or indirectly) tied into some larger web of social connections by
community involvement.
Personal Web Context as a Model of the User:
Role of Communities examined