12User Modeling1

1
Lecture 12 User Modeling

• Topics
• Basics
• Example User Model
• Construction of User Models
• Updating of User Models
• Applications

2
Basics

• User preference vs profile vs model
• A user model is a specification of user
  characteristics aiming to facilitate reasoning
  about his needs, preference and behavior.
• User characteristics include background, mental
  states, interests, interaction patterns, etc.
• Modelling methods
• Knowledge-based approach
• Machine learning approach

3
Basics

• Knowledge-based approach
• Explicitly express user (or user group)
  characteristics in KB in terms of formal KR,
  e.g., FOL, rules, etc.
• Knowledge acquisition (KA) questionnaire,
  interview, observation.
• Reasoning Use KB to reason about user needs or
  difficulties
• Characteristics of knowledge-based approach
• Formal representation
• Reasoning capability

4
Basics

• Characteristics of knowledge-based approach
• KA limitation Hard to comprehend complete and
  consistent user characteristics
• Model updating is equivalent to KB evolution
  very hard to handle concept drift problem

5
Basics

• Machine learning approach
• Learn user characteristics from user behavior
  including user interaction patterns, user
  feedback, etc.
• Example learning mechanisms
• KNN (K-Nearest Neighbor)/ K-Means learn clusters
  based on similarity of vector spaces
• Decision tree learn classification rules based
  on user reviewed solutions and information gain

6
Basics

• Example learning mechanisms
• Naïve Bayesian construct a Bayesian classifier
  based on Bayesian rule according to categorized
  user feedback on proposed solutions
• Bayesian network construct a Bayesian network to
  represent relationships among users actions,
  goals, and system events/states.
• CBR construct a case library to support
  solution prediction

7
Basics

• Characteristics of machine learning approach
• Need high-quality training data
• Need labeled data (from user feedback) if using
  classification techniques
• Model updating is easier but hard to main
  intricate balance between long term interests
  drift and short term interests drift
• High time complexity for on-line processing

8
Example User Model

• Six categories of user characteristics

9
Example User Model

• Example information of Background Knowledge and
  User Idiosyncrasy

10
Example User Model

• Example information of Interaction Preference

11
Example User Model

• Example information of Solution Presentation

12
Example User Model

• Example information of User Interests

• Explicit user feedback
• Interesting degree
• Comprehension degree
• Satisfaction degree
• Definite/most/average/some/none

• Query history
• Solution visit history
• Query time/ solution visit time/ visit sequence/
  ..
• Hyperlinks visit history

13
Construction of User Models

• User Stereotype
• Collect existing user models and cluster them
  into several groups according to the six
  categories of user characteristics
• Define a specification for each group, working as
  the stereotype for the user group
• Or Experts hand-code stereotypes
• Collaborative user modeling
• Fast initialization of a user model for a new user

14
Construction of User Models

• How to do collaborative user modeling
• Get new users basic information through a simple
  questionnaire session
• Cluster the user into one of the user groups
• Use the corresponding stereotype as his initial
  user model
• Update user stereotypes after a sufficient number
  of user models are updated

15
Construction of User Models

• Expert-group stereotype

16
Updating of User Models

• Basic concepts
• Query session (QS)
• From query posted up to feedback returned
• Interaction session (IS)
• From user login up to logout
• Updating of Background Knowledge
• Update Domain Proficiency Table according to
  explicit user comprehension feedback and concept
  difficulty degree as recorded in domain ontology
  (QS)
• Learn user interests from Implicit User Interests
  (several ISs)

17
Updating of User Models

• Updating of User Idiosyncrasy
• Update Terminology Table by analyzing
  user-preferred terms in a given query (QS)
• Updating of Interaction Preference
• Update each query mode according to the user
  interaction pattern (QS)
• Update each recommendation mode according to the
  FAQ-Selection History (QS)
• Updating of Solution Presentation
• Update each presentation mode and corresponding
  presentation ratio according to the FAQ-selection
  history (QS)

18
Updating of User Models

• Updating of User Interests
• Record returned user evaluation in Explicit User
  Feedback (QS)
• Record the user interaction information in
  Implicit User Interests (QS)
• Updating of user stereotypes
• Calculate a statistic (e.g., average) for each
  user characteristic from all user models
• Redistribute user models to user stereotypes
  according to the new statistics
• Recalculate representative values in each user
  stereotype

19
Applications

• Query processing
• User intention extraction according to user
  interests
• Query extension with user interests
• Agent-based computing
• Task delegation, comprehension and processing
• Trust development
• E-learning
• Construction of student model
• E-commerce
• Trust development