Representing User Information Context with Ontologies

1
Representing User Information Context with
Ontologies

• Mohammad Shakil

2
INDEX

• Introduction
• Representing and Maintaining User Context
• Maintaining User Profiles
• Utilizing User Context for Web Search
• Conclusion and Outlook

3
Introduction

• What you need to build Users Information Context
  
• Semantic knowledge about domain being
  investigated
  
• Information expressed in query
• User Profiles
• Paper proposes the Framework which captures
  users
• context via nodes in a concept lattice
  induced from original Ontology and it is updated
  based on the users interaction

4

Problem Consider word Python -gt Web search
engine retrieves all pages containing word
Python -gt Python as a snake, a programming
language or a type of the music -gt System does
not have a knowledge about who is asking for
information and for what
purpose so it gives general answers -gt One size
fits all absence of user preference, search
context or task context
5
Solution

• Contextual Retrieval
• Combine search technologies and knowledge
  about query and user context into a single
  framework in order to provide the most
  appropriate answer for a users information needs

6
How to integrate the essential elements of the
users context?

• Imagine a system where user interacts with a
  domain-specific ontology, represented as a
  concept hierarchy
• Disambiguate the user context using domain
  knowledge inherent in the ontology
• Represent context as an extension of concept
  hierarchy and maintained and updated overtime for
  user profiling
• Use the context developed above in future for
  different information access activity

7
Example
8
Representing User Context

• Domain Ontology Key Concept
• What is Ontology?
• is-a relationship between set of
• concepts thus Organized as a concept
• taxonomy
• Term vector Representation
• Use of documents indexed under concepts
• Compute a weighted term vector for each concept
  along with sub concepts
• Thus aggregation of term vector provides a
  natural partial order among concepts in the
  ontology

9
How the User profile is built?

• User interacts with the presented ontological
  output by selecting and deselecting nodes
• Selection is called Positive evidence
• Deselecting is called Negative evidence
• Profiling attempts to learn user behaviour
  through users browsing and searching
• Use of Concept Lattice

10
Concept Lattice

• User context is represented as a pair of
  elements ltP,Ngt
• When multiple operations are selected then
  positive aggregation is done through min
  operation and negative evidence is done through
  max operation
• Max Greatest Lower bound meaning intersection
  of all vectors presenting in query
• Min Least upper bound meaning union of all
  remaining vectors
• Example. If user give two words in a query then
  through the pages retrieved by these includes
  Positive Evidence which includes only those terms
  appearing at both the place

11
Issues in Maintaining Users Profile

• Once the profile is created it should be
  maintained and updated, a new context should be
  added if necessary
• How this is going to help?
• User Profiles are utilized to provide the
  user with a domain ontology that is more
  consistent with their view of the world
• But this should not be done every time

12
Solution to Updating of User Context

• Build short term context every time for PE and NE
• Compare it to the users long term information
  context
• If similarity between two term vectors exceeds a
  certain threshold then profile should be updated
• Another option is associate each context in user
  profile with specific ontology so user can switch
  between representation of different domain
  ontologies

13
Utilizing User Context for Web Search

• Now it is time to look at how we will use the
  user context to enhance users initial query.
• ARCH (Adaptive Retrieval based on Concept
  Hierarchies) Web Agent
• Processing of a query
• example Python

14
(No Transcript)
15
Just to Refresh 585

• RECALL is the ratio of the number of relevant
  records retrieved to the total number of relevant
  records in the database. It is usually expressed
  as a percentage.
• PRECISION is the ratio of the number of relevant
  records retrieved to the total number of
  irrelevant and relevant records retrieved. It is
  usually expressed as a percentage.

16
Analysis
17
Conclusion

• We discussed framework for Contextual information
  access using ontologies
• Use of long term user profile
• Method of capturing users context via nodes in
  concept lattice and updating the user context
  upon users interaction
• Thus this can definitely improve effectiveness of
  the search queries
• Future work involves modeling the framework to
  include long term vector representation and
  defining object oriented representation of term
  vector so that relationship between different
  concepts can be established according to their
  properties.