Download Interception of User`s Interests on the Web

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Interception of
User’s Interests
on the Web
Michal Barla
Supervisor: prof. Mária Bieliková
[email protected]
Motivation
• Adaptation is based on user model
• Manual filling of user model brings several issues
– Not what a user really wants to do
– User may over/under estimate herself
– User may not know exactly some needed
characteristics
• Goal: Estimate user characteristics automatically
by analyzing user behavior within a system
DC AH 2006
Interception of User's Interests on
the Web
2
Process
Data about user
Collects
User Modeling
Processes
System
User Model
Processes
Adaptation
Adaptation effect
Peter Brusilovsky. Methods and techniques of adaptive hypermedia.
User Modeling and User-Adapted Interaction, 6(2-3):87–129, 1996.
DC AH 2006
Interception of User's Interests on
the Web
3
Data Collection
• We create logs of user activities
• Usually done on server side
– Advantage: always available
– Issue: server is not aware of all performed actions
• Back button and browser cache
• Active elements on a page – e.g. hover
• Monitoring on client side
– Use of client web technologies (JavaScript, Java applets)
– Advantage: we can capture all actions with exact timestamps
– Issue: we have no control on execution of logging tool
DC AH 2006
Interception of User's Interests on
the Web
4
Data Collection – our approach
• Combination of serverside and client-side
logging
Client
Click
XmlHTTPRequest/SOAP
• Client Side Action
Recorder – Click
Presentation layer
– Monitoring on client side
• SemanticLog
SemanticLog
SOAP
Presentation
tools
Log of user’s activity
– Specialized server side
logging tool
DC AH 2006
HTTP request/response
Other layers of the system
Interception of User's Interests on
the Web
5
Click
• Based on JavaScript
– Native access to DOM
• Captures events fired by browser
– Load, Unload, Click, Mouseover,…
• For each event, it records
– Type of event
– Timestamp
– Event context (e.g. what link was pressed)
• Event handling based on W3C DOM Level 2 Event Specification
– Easy integration into existing static pages and dynamic pages
• Communication with server is done asynchronously using AJAX
DC AH 2006
Interception of User's Interests on
the Web
6
Data analysis - challenge
• No direct connection between user behavior and user
characteristics
• User may behave in contrast with her characteristics
(which also characterize such user)
• People are changing  characteristics are changing
• Goal: Estimate characteristics
– Each characteristic has some confidence
– Not all possible characteristics (suitable for a set of domain
characteristics)
DC AH 2006
Interception of User's Interests on
the Web
7
Data analysis - approaches
• Analysis of navigation
– What path did user choose to reach desired information?
• Analysis of user feedback
– Explicit or implicit
– What are the reasons of different ratings?
• Analysis of consistent behavior
– Does user behave according to previous sessions?
– Is user model still valid?
DC AH 2006
Interception of User's Interests on
the Web
8
Analysis of navigation
Real Usage
Data
Contain
Implies
Navigation
Model
Defines goal for
Application
Domain
Usage Patterns
Are evaluated according to
Heuristics
Estimate attributes of
Instance of the
User Model
DC AH 2006
Interception of User's Interests on
the Web
9
Usage patterns identification
• Usage patterns
– pre-defined according to navigation model of a web
site
• Pattern lookup ~ sequence matching
• Usage data
– Stored in a suffix tree structure
• Suffix trie = compressed trie (from “retrieval”)
DC AH 2006
Interception of User's Interests on
the Web
10
Analysis of user feedback
• Searching for implicit feedback patterns on
information objects
– Selection, duration, retention
• Evaluation of feedback  rating
• Rating does not give user characteristics
– Why user rated A differently than B?
– Why user rated P same as Q?
– We get characteristics by comparing concepts
DC AH 2006
Interception of User's Interests on
the Web
11
Concept Comparing - example
• Two similar job offers differs only in duty location
• They get different rating
 We can raise the relevance of duty location
characteristic. We can also estimate desired/unwanted
values.
• Two different job offers have the same rating
 If we find some common aspect, we can raise relevance
of appropriate characteristic
DC AH 2006
Interception of User's Interests on
the Web
12
Analysis of consistent behavior
• Sequential patterns mining on previous user sessions
– Typical user behavior
– Actual session should be mapped to some pattern
• Reasons of inconsistent behavior
– User is in “special” mood, does not follow presumed goal
– User is looking for information on behalf of somebody else
– User has changed
• It has been a long time from previous session
• We invalidate the model and start over
DC AH 2006
Interception of User's Interests on
the Web
13
LogAnalyzer
ConceptComparer
Database
Inconsistent
behavior
detection
Feedback
evaluator
Data
preprocessing
Domain
ontology
User
Model
Updater
Pattern
detection
Log Analyzer
Logs
DC AH 2006
Behavior
patterns
User
model
Heuristics
Interception of User's Interests on
the Web
14
Conclusions
• User modeling based on user behavior analysis
• Acquiring of activity logs on client and server side
– JavaScript based logging tool – Click
• Various approaches to analysis of acquired logs
– Navigation
– Feedback ~ Concept Comparing
– LogAnalyzer
DC AH 2006
Interception of User's Interests on
the Web
15