Download - Personalized Interaction With Semantic Information Portals Eric Schwarzkopf DFKI ([email protected])
Motivation & Goals (1)
Literature research consists not only of information retrieval, but includes also
● Making sense of retrieved information● Locally organizing this information● Sharing gained knowledge
But existing portals focus on the IR process.
Motivation & Goals (2)
● Support additional research activities● Identify and utilize relationships between
interactions to assist the user
– What are the relationships between sensemaking and information management?
– How can those be used to assist in information retrieval?
Interaction Scenario (1 / 3)
Document Collection
1. build
s up
Working Context
2. uses
Information Portal3.1 expresses recommendation need
3.2 evaluates
User ModelRecommendations
3.3 updates3.4 ge
nerat
es
Interaction Scenario (2 / 3)
1. browses
Information Portal
User Model
Recommendations
3.2 updates
Interaction History
2. implicitely creates 3.1 evaluates
Current Location
3.3 adds recommendations4. evaluates options
Interaction Scenario (3 / 3)
1. identifies relevant items
2. adds items
Current Location
3. annotates items
Document Collection
Semantic Relationships
Conceptual Overview
Information Portal
Document Manager
Recommender
Client
Server
•Browsing•Searching
•Organizing•Sensemaking
•Expressing user'sknowledge and needs
•Requestingrecommendations
•Utilizing IR-functionality
Document Manager: Requirements
What are the main features a document manager has to provide?
– Possibility to spatially organize documents.
– Easy creation of informal structures.
– Creation of hierarchical structures.
– Annotation of documents.
– Possibility to define and browse semantic relationships.
– Searching the document collection.
– Sharing of created structures.
Document Manager: WhiteboardCentral interaction component: A 2d whiteboard.
– Allows spatial organization of documents.
– Facilitates creation of informal structures and hierarchies.
Document Manager: AnnotationsDocuments can be associated with free-form text annotations...
– Intended for pointing out relevant information contained in the document.
...and short descriptions displayed on the whiteboard.
– Intended as brief reminders.
Document Manager: Semantic Relationships (1 / 5)
What about relationships between documents?
– “X contains a nice example of the material presented in Y.”
– “X improves on Y.”
– ...
Spatial organization and group hierarchies are not appropriate...
– Clutters the whiteboard.
– Not possible to express detailed semantics.
– Only facilitates the definition of symmetric relationships.
Document Manager: Semantic Relationships (2 / 5)
...nor are notes or short descriptions.
– Cumbersome to define and utilize relationships.
– Relationships are hidden from the system, so it cannot use them to provide any assistance to the user.
Our approach:
– Provide an explicit, internal representation of semantic relationships.
– Allow the user to easily create and modify this representation via the whiteboard.
– Facilitate browsing this semantic network.
Document Manager: Semantic Relationships (3 / 5)
Definition of relationships via drag and drop.
Browsing via a graph-based interface.
Document Manager: Semantic Relationships (4 / 5)
User creates a semantic structure on top of the set of documents and the associated general metadata.
● Semantic relationships
– Content-based relationships
– Problem-specific relationships
● General metadata
– Bibliographic data
– Taxonomic relationships
● Document set
– unstructured
Document Manager: Semantic Relationships (5 / 5)
Further applications of the semantic network:
– Query Augmentation● Identify documents related to the user's current query.● Use IR query augmentation techniques on this set of documents.
– User Modeling● Characterize current working context.● Assess interests and knowledge.
– Recommender● Provide relationship-specific recommendations.● “The work described in this document improves on the work presented in the
document you are currently reading.”
Document Manager: Searching & Sharing
Support for text search.
– Full contents of documents.
– Notes associated with documents.
– Bibliographic data.
Collaborative support limited to sharing whiteboards.
Recommender
Will provide content and strategic recommendations:
● Suggest papers likely of interest
● Recommend concepts to browse or limit a query to
– Names of authors
– Topic areas
– Conferences
● Provide hints for specializing or generalizing queries
Recommender: Components
Recommender
Document Repository
Information Sources:● Knowledge Base● Semantic Networks● ...
Strategies
How to use info sources given a recommendation need.
Recommender: Semantic Networks
Example Query: What documents use material presented in a given document?
Partial Net Partial Net Partial Net
Global, Weighted Network
Combine nets created by the user community.
Weight edges accordingto confidence.
User Model
The UM represents:
● The user's knowledge and interests.
● The user's current working context.
● The user's current goal.
UM requirements:
● Intelligible.
● Accurate.
● Quickly adaptable.
● Comparable, mappable to KB and document space, ...
User Model: Conception
Proposed conceptual model consists of three layers:
– General, fairly accurate, slowly adapting long-term profile.● Represents general interests in terms of ontological concepts.
– More specific, medium-term profile.● Represents the current working context (ontological concepts, characteristic
terms).
– Very specific, quickly adapting, but likely less accurate short-term profile.
● Represents the user's current information need (ontological concepts, characteristic terms).
User Model: Data Sources
General Knowledge
Working Context
Current Information NeedInteraction withInformation Portal
● Recently accessed pages● Recent queries
Interaction withDocument Manager
● Recently accessed documents● Recent modifications of the KB
Personal KB
● Contained items● Structure
UM Layers
User Model: Usage
UM
Data Functions
Query Augmentation
Result Filtering
Recommendations
Contains sets of weighted documents + metadata