Contextbased Search in Topic Centered Digital Repositories

1
Context-based Search in Topic Centered Digital
Repositories

• Christo Dichev, Darina Dicheva
• Winston-Salem State University
• Winston-Salem, N.C. USA
• dichevc, dichevad_at_wssu.ed

2
Topic Maps

• Model for Organizing and Locating Information
• With a distinctive feature to support
• Single access point to all relevant information
  about certin topic.
• Derived features
• Aids both finding the right and finding the
  related information.
• Provides a ground for exploratory search.
• Can we do more in e-learning applications?
• Possible approach - incorporate a contextual
  framework to improve exploration experience.

3
Topicality is not everything

• What kind of things the users are searching for ?
• A typical use of e-learning repositories is for
  locating resources related to certain task.
• searching for topics is only one aspect of the
  search.
• Besides the topic users need resources satisfying
  additional criteria, for example
• Beginner level code examples on Prolog
• Advanced articles related to Prolog negation.
• Which subtopics of a more general topic are
  related.
• Are backtracking and negation related ?
• What about negation and closed world assumption?
• The task requires resources from different
  subtopics.
• writing a paper on Programming Techniques

4
Support for exploratory search

• How to support exploration ?
• Provide a cue where the most promising area for
  exploration lies.
• A possible cue
• Set of documents partially satisfying users
  criteria for relevancy.
• Assumptions
• The targeted resources can be described by
  specifying a locality in the TM, coupled with
  contextual factors for additional filtering.
• The locality can be expressed in terms of
  topics and relations paired with a traversal
  mechanism that identifies what is in and what is
  out of the locality.
• The retrieved set, can be filtered based on the
  remaining criteria.

5
The Approach

• We explore this idea in two directions
• Defining expressions that enables users to
  specify context-based queries
• Using the result of the context-based queries for
  improved query-browsing interaction.

6
.And Now.

• Commercials .

7
TM4L - Topic Maps For E-learning

• TM4L - intended to complement existing Topic Map
  editors and visualization tools.
• It combines
• TM4L Editor
• TM4L Viewer
• Two groups of users are targeted
• authors with a limited background of ontologies
• learners seeking information support in their
  course tasks.
• TM4L is currently available as a standalone
  application.
• http//compsci.wssu.edu/iis/nsdl/download.html

8
TM4L Functionality

• In TM4L the initial set of relations includes
  five relations
• Whole-Part
• Superclass-Subclass
• Instance-Of
• Related
• Similar
• This set can be extended with arbitrary user
  defined relations.

9
TM4L Partonomy View
10
Topic Search in TM4L
11
Example of TM4L Application
12
Contextual Aspects in Information Organization

• In practice, we group topics and resources based
  on a certain set of relation types.
• It allows user to select from all related topics
  the ones that are related in a certain way.
• This suggests the following strategy
• Users specify the relation types they are
  interested in combined with some contextual
  factors
• The system draws the locality along the user
  defined axes, filtering the resources within the
  locality based on contextual factors.
• So resolving the locality is equal to traversal

13
Traversal is Insufficient

• Topic locality traversals enable users to specify
  TM projections and define regions in such
  projections
• by specifying the topics and relation types they
  are interested in
• by traversing the relations to extract the info
  within the region.
• ... but this strategy addresses only topicality
• Topic and resource grouping may be done on
  several ontological levels that reflect different
  contexts.
• When users search for resources, several criteria
  are in play
• What is the resource about ?
• In what form and level is the content presented?
• How is it related to the current users task?

14
TM4L Perspective

• In our context-based IR framework, topic-relation
  traversal is intended to capture the aboutness
  of the search.
• Some other contextual factors
• Level of difficulty such as, depth of coverage
  technicality etc. are captured by the
  Scope/Theme property.
• Instructional factors such as Lecture Notes,
  Examples, Tutorial, Exercise, Demo etc. are
  captured by Resource type property.

15
Contextual Queries in TM4L
16
Similar vs. Relevant

• In information retrieval, the terms similar and
  relevant are often used interchangeably.
• Concepts are related, and their relationships
  imply some kind of similarity.
• This aspect of similarity is addressed by the
  contextual queries, which
• restrict the query within a user defined region
• filter the result on the basis of document
  properties.

17
Different Aspects of Similarity

• Resources can be similar
• by topic,
• by the level of granularity,
• by the source,
• by the presentation style, etc.
• Assumptions
• Similar resources can be shelved together.
• A resource can be similar in different aspects to
  some other resources
• A resource can be in different kind of shelves
  at the same time
• Retrieved resources, can point to their (multiple
  kind of) shelves, where the user can find other
  relevant material.

18
Context is Elusive

• Some intuition
• While It is not always possible to articulate
  contexts, we are able to recognize items matching
  our context
• Possible approach
• If you see item matching your context then
• find similar items

19
Defining a Context in TM4L