Download - Modelling Personalization
I’m not youModelling and conceptualizing personalization in information architecture
Presented by / Bogo Vatovec
Change Management / Knowledge Engineering / User Experience /Interaction Design / Process Engineering
© bovacon 2008
Integration of personalization in a design of an interactive system.
What will I be talking about?
© bovacon 2008
Personalization related deliverables
» Personalization Strategy
» Introduction to personalization
» Writing personalization scenarios
» Selecting the right strategy for my project
» Goals and KPIs
» Continuous Improvement program
» Conceptual models
» User, content and functional profiles
» Aspect-oriented use cases
» Personalization rules
» Probabilistic models
» Technology Selection
» Architectural requirements
» Choosing the tool
» Testing and Acceptance
I‘m going to skip the red ones….
© bovacon 2008
Introduction to personalization
© bovacon 2008
Personalization can mean lots of different things to different stakeholders
© bovacon 2008
Some of these cases are trivial
If you are happy enough that they apply to you, forget the rest of the presentation.
For the others…
© bovacon 2008
First some categorization
No pretention that this modell is complete
© bovacon 2008
Introduction – how personalization works
Personalization works essentially using the following methods:
1. Explicit settings: I want A, B and C.2. Similarity of profiles: If A is similar to B then do C.3. Sequence of actions: If 1) and 2) and 3), then do 4).4. Combination of: If A is similar to B and 1) and 2) then do 4)
© bovacon 2008
Explicit personalization – the user chooses options
© bovacon 2008
Implicit personalization
© bovacon 2008
Most popular – collaborative filtering
If user A is looking at product 1 and has a similar profile to user B and user B also bought product 1 and bought product 2, recommend product 2 to user A.
User 1 User profile 1
Content A
has
then recommend
readsUser profile 2read
Content B
Also read
Example: collaborative filtering
© bovacon 2008
Conceptualizing personalization
© bovacon 2008
Conceptualizing personalization has one basic rule
Describe behaviour without and with personalization separately.
© bovacon 2008
Separate behaviour without and with personalization
Without personalization,
the user experience looks like this…..
With personalization,
the user experience looks like that…..
Personalization expands the existing essential functionality. It is never a function on its own.
The interactive system must serve its main purpose also without personalization.
© bovacon 2008
IAs use common usage scenarios to define the context of personalization
© bovacon 2008
Make sure to consider the following
» Align personalization with the business strategy and desired user experience
» Is it important to have user profiles?
» Which data about the users is important?
» What kind of a benefit does the interactive system provide for the user and the business through different personalization types?
» Clearly state the personalization strategy by:
» Describing the personalization type used
» Describing usage scenarios illustrating the experience
» State the benefits for the business and the user
» Make sure every stakeholder understands this
© bovacon 2008
Visualizing the big picture
© bovacon 2008
Conceptual models - a „rich picture“
© bovacon 2008
Conceptual models –A relationships diagram
Behavioural trackingengine
User
experience
environment
Influences behaviour
Influences possibilities
Website
Uses
Core profile
Extended profile
Personalization engine
Content profileFills out
monitors
Fills out
Generates pages
Function profile
compares
© bovacon 2008
Profiles, profiles, profiles
© bovacon 2008
User profiles
A0: Anonymous first time visitor
A1: Anonymous returning visitor leaving traces
R1: Registered user (essential profile)
R2: Registered returning user (essential and extended profile)
RP1: Registered returning user with behaviour (essential, extended and behavioural profile)
User profile ID
Behavioral data
Core profile
Extended profile
A0
A1
R1
R2
RP1
© bovacon 2008
Content and function profile
Content/function
Attributes
Tags
Tag cloud
© bovacon 2008
Example: Detailing the user profile
© bovacon 2008
Clarify user profile management
» When is the user profile created?
» When is the user profile deleted?
» Can the user delete his profile?
» What happens to the content related to the user profile?
» How much can the user know about his profile?
» How can the user manage/modify his profile?
© bovacon 2008
It‘s time for use cases and flowcharts
© bovacon 2008
Use cases in Information Architecture
Illustrative purpose only
Learn about banking products
Open an account
User System
U1
S1
S2
U2
Screen 1
Screen 2
© bovacon 2008
Use case model allows us to better understand the interactive system
Open an account
Open a daily money account
Open a term deposit account
Open a giro account
generalize include
Open an account
Provide personal
information
Provide bank details
Provide transaction
details
Include
Include
Include
A use case model consists of three main type of relationships:
Illustrative purpose only
© bovacon 2008
The aspects-oriented use case modelling gives extend relationship a dynamic dimension
extend
Open an account
User logged in
User has an account already
Apply business rules
extend
extend
extend
Illustrative purpose only
© bovacon 2008
This dynamic dimension changes system behaviour
Open an account
User System
U1
S1
U2
Screen 3
User has an account already
extend
© bovacon 2008
User profiles can be incorporated in the activity diagram
User System/A1
U1
S1
U2
Screen 3
System/RP1
S1
S2
S3
Screen 1
Screen 2
For use case haters:Use the flow-diagramm connecting screens in the same way and create various flow for each profile-type
© bovacon 2008
This is silly.Have no idea anymore what we are talking about
Scenario Use cases Activity steps User profileConsists of Consists of Affected by
Behavioural tracking
Generated by
user
Generated by
© bovacon 2008
Bringing it all together in a readable form
© bovacon 2008
Modelling using aspects-orientation provides several advantages
» Makes conceptualization and incremental detailing easier
» Enables iterative development
» Allows for functional reduction
» Allows for easier testing and verification
© bovacon 2008
This was easy, wasn‘t it?
Wait – were are at the personalization rules?Wait – were are the Bayesian models?Wait – were are adaptive interfaces?
© bovacon 2008
Personalization rules specify conditions when personalization applies
Some examples
1. If user is looking at content A and content B has similar tags, show content B as related content.
2. If user A is looking at product 1 and has a similar profile to user B and user B also bought product 1 and bought product 2, recommend product 2 to user A.
3. If user A with profile A has click 1 then 2 then 3 and pauses for 2,5s here, show context sensitive help.
» In the conceptualizing process, we use a simple descriptive language like this.
Notice example Number 2) – it has a sequence of conditions…. Will talk about this in a minute.
© bovacon 2008
Let‘s move to technology a bit and then explore the rest
© bovacon 2008
Personalization strategy defines the use cases for technology
» Capture user behaviour on the website and store it in the user profile.
» Dynamically generate content recommendations based on user behaviour in real time (in one session)
» Dynamically generate content recommendations based on the stored user profile.
» Provide for statistical analysis of user behaviour based on user profiles to expand marketing knowledge.
» Provide the user the possibilities to customize the layout of the home pages.
© bovacon 2008
To come up with requirements, you need to further specify them
© bovacon 2008
Technical architecture for personalization
Users
User interface Layer Personalization
Profile Layer Specific Values
Vocabulary Layer Attributes
Content
User Profile
Content Profile
Content Attributes
User Attributes
Personalization Rules
Personalization Rules
Modified from the model by Argus Center for Information Architecture
AnalyticsUser
BehaviourContent Statistics
© bovacon 2008
How can this model help us?
» How much does one system require explicit user profile setting?
» How much implicit?
» How does it integrate with other systems to fill out the profiles?
» How does the system support vocabulary level personalization rules?
» How easy it is to automatically populate content values?
» What features does the personalization system include to manage controlled vocabularies?
© bovacon 2008
Those of you who carefully listened notices that I missed an important compontent…..
TIME
© bovacon 2008
Until now we looked at the user and content profiles as one object with one state
Open an account
User System
U1
S1
U2
User profileFills out
Fills out
Fills out
Since the user profile has only one state, we forget the sequence of actions – user‘s decision making process
© bovacon 2008
Here the fun begins: Adaptive user interfaces and even – yes – artificial intelligence
Adaptive user interfaces are interfaces which automatically acquire knowledge about the users, update this knowledge over time, and uses this knowledge to adapt to the users’ requirements.
Artificial intelligence is defined as as "the study and design of intelligent agents, where an intelligent agent is a system that perceives its environment and takes actions which maximize its chances of success.
© bovacon 2008
Remember the personalization rule from before
If user A with profile A has click 1 then 2 then 3 and pauses for 2,5s here, show context sensitive help.
To model this, we need to have a user profile with a sequence of events
Open an account
User System
U1
S1
U2
User profile 1'
Fills out
Fills out
Fills out
User profile 1'’
User profile 1'’’
© bovacon 2008
» How can we track all past user actions? Which data do we need to track?
» Can previous behaviour give us enough information for action?
» How can we provide qualitative personalization rules based on this data?
» Can previous behaviour provide us enough information to predict future user actions?
A Wizard of Oz research methodology in reverse – can the experts predict user action?
We talk about probability, game theory and statistics…
Considering time brings a new dimension of complexity and challenges
© bovacon 2008
This is where probability theories and machine learning come into the game
The good news is that – unless you have a peculiar interest in this – you can get along with the first part of the presentation.
http://en.wikipedia.org/wiki/Machine_learning
© bovacon 2008
Summary – remember what we talked about…
» We looked at the following topics
» Personalization types and categorization
» Personalization scenarios
» User, content and functional profiles
» Aspect-oriented use cases
» Personalization rules
» Probabilistic models
» Architectural requirements
© bovacon 2008
Summary – the doggy bag
» What you should take with you
» Personalization was not invented by Amazon.
» Define what personalization is for you
» Explain this to every stakeholder
» Model personalization so that project members understand how it will work
» Use various models for various stakeholders
» Explore various modelling methods and start using them. They help in the communication and clarification of issues and open points.
Thank you!
Bogo Vatovecbovacon
Boxhagener Str. 111 / 10245 BerlinT +49 30 20078666 / F +49 30 20078661 / [email protected] / www.bovacon.com
© bovacon