Seminar: Web Mining

1
Seminar Web Mining

• Web Mining and Personalization
• A Business Perspective
• Presented by Dejan Vasiljev

2
Overview

• Introduction What why
• The phases of personalization
• Web mining and personalization the relationship
• Conclusions

3
Personalization defined

• Web personalization is driven by computer which
  tries to serve up individualized pages to the
  user based on some form of model of that users
  needs (Nielsen, J.)
• Web personalization can be described, as any
  action that makes the Web experience of a user
  personalized to the users taste (Mobasher et
  al.)
• Essentially, trying to infer what the user may
  need or want and adjusting the offer accordingly

4
Why personalize?

• Claim Personalization is not a trend, but a
  necessity!
• The conditions in which firms do business have
  changed dramatically (new technologies,
  internet)
• So, the ways of production had to change as well
  from mass production to mass customization
• From Ford T to Amazon.com

5
Why personalize?

• Find out as much as possible about your
  customers, their characteristics, preferences,
  habits and interests
• Use that information to personalize your
  product/service and create value for your
  customers gt capture higher profits
• Treat you customer as an individual, because if
  you dont somebody else will!

6
A common misperception Personalization v.
customization

• Personalization is driven by the computer which
  tries to serve individualized pages to the user
  based on some form of model of that users needs
• Customization is under direct user control the
  user explicitly selects between certain options
• Personalization example amazon.com
• Customization example my.yahoo.com

7
Personalization the phases

• Personalization process can be broken into three
  essential phases
• Collecting customer information
• Analyzing information and generating insight
• Applying the generated insights

8
Phase 1 Collecting customer information

• Customer data can be obtained in different ways
• some can be observed by the system directly
• some require additional steps
• In general, there are two types of customer data
  
• user data information about personal
  characteristics of the user
• usage data information about users interactive
  behavior
• The obtained data is not always reliable

9
Phase 1 (Collecting information) User data

• Companies collect different type of data, that
  can help infer users characteristics and
  personalize their offer
• demographic data
• data about user knowledge, skills and
  capabilities
• data about user interests and preferences (e.g.
  adjust the promotion of cars to different
  audiences)
• data about user goals and plans (e.g. information
  or product)

10
Phase 1 (Collecting information)Usage data

• Usage data can be acquired by observing and/or
  analyzing users interactive behavior
• So, what should be observed?
• Answer
• selective actions of users (such as clicking on a
  link) can reveal a lot about interests,
  unfamiliarity with various technical terms,
  preferences
• temporal behavior
• ratings (binary, rating scale)
• purchases and purchase-related actions (e.g.
  Amazon)

11
Phase 1 (Collecting information)Acquisition
methods

• Two major groups
• user model acquisition methods for acquiring
  explicit assumptions about user data
• user-supplied information
• passive acquisition
• plan recognition
• stereotype reasoning
• usage model acquisition methods to obtain
  information about user behavior
• correlations between situations and actions (e.g.
  Microsofts personal assistant)
• actions sequencing

12
Phase 2Analysis and insight generation

• After the data collection, process the data
  further in order to generate useful insights
• The following techniques are used
• Deductive reasoning - from the more general to
  more specific cases
• Inductive reasoning - from specific cases to the
  general case
• Analogical reasoning from similar cases to the
  present case

13
Phase 2 (Data analysis)Deductive reasoning

• The assumption that the user knows concept X is
  represented by entering a representation of that
  concept into the user modeling knowledge base
• Once added, this assumption can trigger further
  meta-level reasoning based on the concept
  relationship
• For example, from dolphins to whales and mammals
  (concept hierarchy)

14
Phase 2 (Data analysis)Inductive reasoning
Learning

• Inductive reasoning is about monitoring the
  users interaction with the system and drawing
  general conclusions based on a series of
  observations
• Helps to learn about the user by using learning
  algorithms and is used for inferring users
  interests
• Feature-based filtering relies on certain
  features of an object of users interest
• E.g. user interest in movies is determined by
  preferences about genre, actors, director..

15
Phase 2 (Data analysis)Analogical reasoning

• Web-based systems have a large number of users
  use analogical reasoning to recognize
  similarities between users
• Two commonly used methods
• Clique-based filtering matching a single
  profile with profiles of similar users
• Clustering forming groups of user profiles

16
Phase 2 (Data analysis)Analogical reasoning

• Clique-based filtering for a given user, the
  system tries to find users with similar
  interaction behavior
• Then the system adapts to the individual user
  based on the behavior of similar users
• This process usually has three steps
• Find similar neighbors
• From the group of similar neighbors, select a
  comparison group that is closest to the user
• Derive predictions

17
Phase 2 (Data analysis)Analogical reasoning

• Clustering user profiles classify the users into
  categories
• Machine learning methods and statistics are used
  to form user profiles
• The system applies a clustering algorithm to find
  similar users and form group profiles
• Information from related group profiles can be
  used if the needed information for the individual
  user is not available

18
Phase 3Application of generated insights

• Use the acquired knowledge for personalization
• Personalize your pages and create value for users
  which will enable you to reap higher profits
  (hopefully)
• Personalize
• Content
• Presentation and media format
• Structure (links)

19
Phase 3 (Application) Personalize your content

• Personalize the content by offering
• optional explanations
• optional detailed information
• personalized recommendations etc.
• And using following techniques
• page variant personalization on the page level
• fragment variants personalization on the
  fragment (paragraph, image, table..) level
• fragment coloring
• adaptive stretch text extend or shrink the text
  by clicking on it

20
Phase 3 (Application)Personalize presentation
modality

• Content of the presentation stays the same (not
  always so), while the format and layout of the
  objects change
• Or, change the modality from images to text,
  from text to audio, from video to still images
• Example AVANTI bases the selection of different
  modalities on the users physical (dis)abilities
• A map of Siena, a city in Italy
• The same page, this time for blind people the
  image has been changed to text

21
Phase 3 (Application)Personalize the structure

• Personalization of structure changes the link
  structure or its presentation to users
• Different techniques are used
• Collateral personalization the by-product of
  content personalization
• Link sorting ranking Web pages (e.g. Google)
• Link annotation use different colors and
  symbols to annotate links (e.g. visited links
  change color, Googles sponsored links..)
• Link hiding and unhiding the link is hidden,
  can be discovered by, for example, different
  cursor shape. Or, the link is unhidden after
  visiting certain pages
• Link removal/addition

22
Conclusions

• The concept of personalization is great, but does
  it really work?
• The answer is not clear, but typical Web
  personalization efforts fail to produce the
  results that match market expectations
  (Forrester, 1999)
• In other words, apart from a couple of examples
  (Amazon.com), personalization, although announced
  as the next big thing, perhaps should be treated
  more skeptically

23
Conclusions

• The reasons behind the failure of
  personalization
• Customers dont want relationships with
  corporations, but with people
• Personalization requires data, which is not
  always easily obtainable
• Personalization understands customers poorly
  (even Amazon doesnt always succeed)
• Personalization is expensive

24
Conclusions

• Moreover, personalization can sometimes be viewed
  as offensive by the users dont stereotype me,
  I know what I want better than the computer
• Finally, the ethical dimension of personalization
  should be taken into consideration
• The issue of collecting data versus privacy
  commands attention what is acceptable, and where
  do we draw the line?