Download Human-Centred Web Adaptation and Personalization: From Theory to Practice PDF

TitleHuman-Centred Web Adaptation and Personalization: From Theory to Practice
Author
LanguageEnglish
File Size8.2 MB
Total Pages358
Table of Contents
                            Foreword
Preface
	Our Standpoint: From Theory to Practice
	For Academia and Business
Acknowledgments
Book Overview
Contents
Abbreviations
Part I: Theory: The Human in the Centre of Web Personalization
	Chapter 1: Personalization in the Digital Era
		1.1 Introduction
		1.2 Rethinking Human-Computer Interaction
			1.2.1 Repositioning the “I” in HCI
			1.2.2 HCI Meets Adaptation and Personalization: Influential Research Disciplines
		1.3 The Need for User-Centred Design in Interactive Computing Systems and Interfaces
		1.4 The Concept of Web Adaptation and Personalization
		1.5 Current Problems and Challenges: An ‘Out-of-the-Box’ Thinking
		1.6 Summary
		References
	Chapter 2: Human Factors in Web Adaptation and Personalization
		2.1 Introduction
		2.2 Human Cognition and Information Processing
			2.2.1 The Role of Human Memory
				2.2.1.1 Sensory Memory
				2.2.1.2 Short-Term Memory and Working Memory
				2.2.1.3 Long-Term Memory
			2.2.2 Visual Perception
			2.2.3 Visual Search
				2.2.3.1 Feature Integration Theory
				2.2.3.2 Guided Search Theory
				2.2.3.3 Decision Integration Hypothesis
			2.2.4 Visual Attention, Speed and Control of Processing
			2.2.5 Learning Styles
			2.2.6 Cognitive Styles
				2.2.6.1 Verbal/Imager Dimension
				2.2.6.2 Wholist/Analyst Dimension
			2.2.7 Elicitation Methods of High-Level and Elementary Cognitive Processes
			2.2.8 Implication of Cognitive Aspects on Adaptation and Personalization
		2.3 Emotions and Learning Process
			2.3.1 Emotion Regulation
				2.3.1.1 The Experiential Level
				2.3.1.2 The Rational Level
			2.3.2 Emotional Arousal
			2.3.3 Methods of Extracting Emotions and Anxiety
			2.3.4 Implications of Anxiety in Adaptive Interactive Environments
		2.4 Summary
		References
Part II: Principles: Web Adaptation and Personalization Processes and Techniques
	Chapter 3: User Modeling
		3.1 Introduction
		3.2 User Modeling Factors for Personalization
			3.2.1 User Information
				3.2.1.1 Knowledge
				3.2.1.2 Interests
				3.2.1.3 Goals
				3.2.1.4 Background
				3.2.1.5 Individual Traits
			3.2.2 Context Information
				3.2.2.1 Platform-Oriented Context Modeling
				3.2.2.2 Location-Oriented Context Modeling
				3.2.2.3 Social-Oriented Context Modeling
		3.3 User Data Collection Methods
			3.3.1 Explicit User Data Collection Methods
			3.3.2 Implicit User Data Collection Methods
		3.4 User Model Generation
			3.4.1 Clustering
			3.4.2 Classification
			3.4.3 Association Discovery
			3.4.4 Sequential Pattern Mining
		3.5 Modeling Human Factors in Interactive Systems
			3.5.1 Identifying Intrinsic Human Factors for Building a Comprehensive User Model
				3.5.1.1 The User Perceptual Preference Characteristics
		3.6 Summary
		References
	Chapter 4: Personalization Categories and Adaptation Technologies
		4.1 Introduction
		4.2 Personalization Categories
			4.2.1 Link Personalization
			4.2.2 Content Personalization
			4.2.3 Personalized Web Search
			4.2.4 Context Personalization
			4.2.5 Authorized Personalization
			4.2.6 Humanized Personalization
		4.3 Adaptation Technologies
			4.3.1 User Customization
			4.3.2 Rule-Based Filtering
			4.3.3 Content-Based Filtering
			4.3.4 Collaborative Filtering
			4.3.5 Web Mining
			4.3.6 Demographic-Based Filtering
			4.3.7 Agent Technology
			4.3.8 Cluster Models
		4.4 Semantic Web Technologies for Adaptation and Personalization Systems
		4.5 Leveraging the Social Web for Adaptation and Personalization
		4.6 Adaptation Effects in User Interfaces
			4.6.1 Adaptive Content Presentation
			4.6.2 Adaptive Navigation Support
		4.7 Web Adaptation and Personalization Systems and Frameworks
			4.7.1 PAC
			4.7.2 PersonaWeb
			4.7.3 Hybreed
			4.7.4 Adaptive Notifications in Virtual Communities
			4.7.5 Smartag
			4.7.6 PRESYDIUM
			4.7.7 PERSONAF
			4.7.8 CTRL
			4.7.9 EKPAIDEION
			4.7.10 AdaptiveWeb
			4.7.11 Knowledge Sea II
			4.7.12 CUMAPH
			4.7.13 mPERSONA
			4.7.14 INSPIRE
			4.7.15 SQL-Tutor
			4.7.16 Proteus
			4.7.17 WBI
			4.7.18 ARCHIMIDES
			4.7.19 TANGOW
			4.7.20 AHA!
			4.7.21 SKILL
			4.7.22 ELM-ART II
			4.7.23 BASAR
			4.7.24 InterBook
		4.8 Summary
		References
	Chapter 5: A Generic Human-Centred Personalization Framework: The Case of mapU
		5.1 Introduction
		5.2 A High-Level Adaptation and Personalization Architecture
		5.3 Conceptual Design of mapU
			5.3.1 Module 1: User Modeling
				5.3.1.1 Cognitive Aptitude Tasks
				5.3.1.2 Metrics for Eliciting Emotional Processing Factors
				5.3.1.3 Web Interaction Metrics
				5.3.1.4 Data Processing
				5.3.1.5 User Classification
				5.3.1.6 User Model
			5.3.2 Module 2: Personalization
				5.3.2.1 Content Management
				5.3.2.2 Adaptation Component
		5.4 Design and Development of mapU
			5.4.1 mapU Web Server
				5.4.1.1 Administration and Management
				5.4.1.2 Front-End User Modeling
				5.4.1.3 Adaptive User Interface
			5.4.2 mapU Back-End
		5.5 Technologies and Languages for the Design and Development of the mapU System
			5.5.1 HTML: HyperText Markup Language
			5.5.2 CSS (Cascade Style Sheets): Giving Style to HTML
			5.5.3 Client-Side Languages
			5.5.4 Server-Side Languages and Frameworks
				5.5.4.1 PHP–PHP: Hypertext Pre-processor
				5.5.4.2 ASP: Active Server Pages
				5.5.4.3 Server-Side Language and Framework Used for the mapU System
			5.5.5 Storing and Retrieving Data
		5.6 Summary
		References
Part III: Practice: A Practical Guide and Empirical Evaluation in Three Distinct Application Areas
	Chapter 6: The E-Learning Case
		6.1 Introduction
			6.1.1 Potential, Limitations and a High-Level Classification of E-Learning Systems
				6.1.1.1 Classification Based on ICT
				6.1.1.2 Classification Based on Educational Technologies
			6.1.2 Context-Aware and Activity-Based Considerations in E-Learning Environments
			6.1.3 The Importance of Adapting and Personalizing E-Learning Environments
		6.2 Design Considerations and Constraints
		6.3 Human-Centred Design Guidelines
			6.3.1 Guidelines for E-Learning Environments
				6.3.1.1 Guideline #1: Textual Representation with a Guided, Holistic Structure and Support Tools
				6.3.1.2 Guideline #2: Textual Representation with Analytic Structure and Learner Control
				6.3.1.3 Guideline #3: Graphical Content with a Guided, Holistic Structure and Support Tools
				6.3.1.4 Guideline #4: Graphical Content with Analytic Structure and Learner Control
				6.3.1.5 Guideline #5: Support for Users with Limited Working Memory Capacity
				6.3.1.6 Guideline #6: Support for Users with High Levels of Anxiety
			6.3.2 Guidelines for M-Learning Environments
				6.3.2.1 Guideline #7: Textual Representation with a Guided, Holistic Structure and Support Tools
				6.3.2.2 Guideline #8: Textual Representation with Analytic Structure and Learner Control
				6.3.2.3 Guideline #9: Graphical Content with a Guided, Holistic Structure and Support Tools
				6.3.2.4 Guideline #10: Graphical Content with Analytic Structure and Learner Control
				6.3.2.5 Guideline #11: Support for Users with Limited Working Memory Capacity
			6.3.3 Adaptation Paradigm in mapU Based on Guidelines
		6.4 Evaluation
			6.4.1 Method of Study 1: Eye-Tracking Study
			6.4.2 Method of Study 2: Personalized E-Learning Study
			6.4.3 Method of Study 3: Personalized M-Learning Study
			6.4.4 Results
				6.4.4.1 Analysis of Study 1: Eye-Tracking
				6.4.4.2 Analysis of Study 2A: Personalization in E-Learning
				6.4.4.3 Analysis of Study 2B: Personalization in E-Learning
				6.4.4.4 Analysis of Study 3: Personalization in M-Learning
		6.5 Benefits, Impact and Limitations
		6.6 Summary
		References
	Chapter 7: The E-Commerce Case
		7.1 Introduction
			7.1.1 Potential and Limitations of Multi-channel E-Commerce Products and Services Delivery
			7.1.2 Why to Adapt and Personalize E-Commerce Environments
		7.2 Design Considerations and Constraints
		7.3 Human-Centred Design Guidelines
			7.3.1 Guidelines for E-Commerce Product Views
				7.3.1.1 Guideline #1: Textual Representation with Holistic Structure and Additional Navigation Support Tools
				7.3.1.2 Guideline #2: Textual Representation with Analytic Structure
				7.3.1.3 Guideline #3: Diagrammatical Representation with Holistic Structure
				7.3.1.4 Guideline #4: Diagrammatical Representation with Analytic Structure
				7.3.1.5 Guideline #5: Additional Support for Users with Limited Working Memory Capacity
			7.3.2 Guidelines for E-Commerce Checkout Process
				7.3.2.1 Guideline #6: Single One-Page Checkout Design
				7.3.2.2 Guideline #7: Step-by-Step Checkout Design
			7.3.3 Adaptation Paradigm in mapU Based on Guidelines
		7.4 Evaluation of Product Views Personalization (Based on Sony Design)
			7.4.1 Methodology
			7.4.2 Results
				7.4.2.1 Task Completion Performance
				7.4.2.2 Task Completion Accuracy
				7.4.2.3 User Satisfaction
		7.5 Evaluation of Product Views Personalization (Based on HP Design)
			7.5.1 Methodology
			7.5.2 Results
				7.5.2.1 User Modeling
				7.5.2.2 Task Performance Between Original and Personalized Design
				7.5.2.3 Users’ Perceived Usability
		7.6 Evaluation of Checkout Process Personalization
			7.6.1 Methodology
			7.6.2 Results
				7.6.2.1 Task Efficiency
				7.6.2.2 User Preference
		7.7 Benefits, Impact and Limitations
		7.8 Summary
		References
	Chapter 8: The Usable Security Case
		8.1 Introduction
			8.1.1 User Authentication
			8.1.2 Human Interaction Proofs (CAPTCHA)
			8.1.3 Why to Adapt and Personalize Security-Related Tasks
		8.2 Design Considerations and Constraints
			8.2.1 Design Considerations in Knowledge-Based User Authentication
			8.2.2 Security Considerations in Knowledge-Based User Authentication
			8.2.3 Design Considerations in CAPTCHA
			8.2.4 Security Considerations in CAPTCHA
		8.3 Human-Centred Design Guidelines
			8.3.1 User Authentication Mechanisms
				8.3.1.1 Guideline #1: Text-Based Password with Standard Complexity
				8.3.1.2 Guideline #2: Text-Based Password with Higher Complexity
				8.3.1.3 Guideline #3: Recognition-Based Graphical Authentication with Standard Complexity
				8.3.1.4 Guideline #4: Recognition-Based Graphical Authentication with Higher Complexity
			8.3.2 CAPTCHA Mechanisms
				8.3.2.1 Guideline #5: Text-Recognition CAPTCHA with Standard Complexity
				8.3.2.2 Guideline #6: Text-Recognition CAPTCHA with Higher Complexity
				8.3.2.3 Guideline #7: Image-Recognition CAPTCHA with Standard Complexity
				8.3.2.4 Guideline #8: Image-Recognition CAPTCHA with Higher Complexity
			8.3.3 Adaptation Paradigm in mapU Based on Guidelines
		8.4 Evaluation
			8.4.1 Study Design Methodology
				8.4.1.1 Phase A: User Modeling
				8.4.1.2 Phase B: Initial User Interactions
				8.4.1.3 Phase C: Swapping the Security Mechanism
				8.4.1.4 Phase D: Post-study Survey
			8.4.2 Participants
			8.4.3 User Interaction Metrics
			8.4.4 Hypotheses
			8.4.5 Analysis of Results
				8.4.5.1 User Modeling
				8.4.5.2 Personalized Versus Non-personalized User Authentication
				8.4.5.3 Personalized vs. Non-personalized CAPTCHA
				8.4.5.4 Post-study Survey
		8.5 Benefits, Impact and Limitations
		8.6 Summary
		References
Epilogue
                        
Document Text Contents
Page 1

Human–Computer Interaction Series

Panagiotis Germanakos
Marios Belk

Human-
Centred Web
Adaptation and
Personalization
From Theory to Practice

Page 2

Human–Computer Interaction Series

Editors-in-chief
Desney Tan , Microsoft Research , USA
Jean Vanderdonckt , Université catholique de Louvain , Belgium

Page 179

154

The linearity of users’ navigation is modeled through the absolute distance of
links (ADL), which is the total absolute distance between the links visited by a user
ui to be:


nav u

x x x

N
adl

i
i

N

i i

( ) =
- + -å

=
-1

2
11



In the aforementioned equation, xi represents the identifier of links visited, i.e.,
i = 1 is the first link visited (x1 is equal to 1), i = 2 the second (x2 is equal to 2) and so
on, and N is the number of total links clicked. Thus the distance between sequential
links is assumed to be equal to 1.

5.3.1.5 User Classification

For classifying users into specific groups (e.g., Verbal or Imager group), two widely
used methods exist in the literature: (i) grouping users based on a predefined thresh-
old value for each psychometric test or questionnaire which is suggested and stan-
dardized by the creator of each inventory (e.g., based on the aforementioned
processed responses of a user (e.g., Verbal/Imager ratio), the user is grouped in a
particular group, given a predefined range of thresholds); and (ii) grouping users
based on cluster analysis which aims to divide a set of users into cluster groups that
are different from each other and whose members are similar to each other accord-
ing to each of the aforementioned processed values. In the context of mapU, we
classify users based on cluster analysis for the following reasons: (i) the suggested
thresholds are standardized and evaluated based on a specific population which
might not be representative for different populations under investigation; and (ii)
cluster analysis could yield very good results in our case since the data being pro-
cessed represent a scale with two end points (e.g., low and high values of the Verbal/
Imager ratio represent respectively Verbal and Imager users), and thus can effec-
tively separate users from each other depending on the responses to each stimuli
(Belk et al. 2013). Apparently, cluster analysis also entails practical limitations,
such as initialization issues, i.e., when a limited number of users are registered in
the system, making the cluster analysis difficult and ineffective to perform.

In this respect, depending on the aforementioned processed values of each user
(i.e., λv : g(ui), λw : a(ui), μsc(ui), μmab(ui), λvt : vg(ui), navadl(ui), stait(ui), bars(ui), cta(ui),
er(ui)), users with close distance values will be grouped in the same cluster.
Accordingly, the following characteristics for each user are finally elicited based on
the cluster the user is assigned: (i) A user is either a Verbal, or an Imager (based on
λv : g(ui) or λvt : vg(ui)); (ii) a user is either a Wholist, or an Analyst (based on λw : a(ui) or
navadl(ui)); (iii) a user has either limited or enhanced speed and control of processing
(cognitive processing efficiency); (iv) a user has either limited or enhanced working
memory capacity; (v) a user is in general less or highly anxious in life; (vi) a user is

5 A Generic Human-Centred Personalization Framework: The Case of mapU

Page 180

155

currently highly anxious or less anxious; (vii) a user is less or highly anxious within
a specific application domain; and (viii) a user is less able or highly able to regulate
his emotions.

For user classification we utilize the k-means clustering algorithm since it is
considered one of the most robust and efficient clustering algorithms (Wu et al.
2007). The k-means clustering algorithm is performed as follows: The algorithm
initially sets the data point with the smallest value (e.g., Verbal/Imager ratio) as the
first cluster centre and the data point with the largest value as the second cluster
centre. Given that the desired groups are known in our case (e.g., Verbal and Imager),
the algorithm is set to k = 2. The distance between all other data points and cluster
centres are then calculated, and each data point is assigned to the cluster whose
distance from the cluster centre is the minimum of all the cluster centres using the
Euclidian distance. New cluster centres are recalculated by measuring the mean of
all data points of each newly created cluster. Next, the distances between each data
point and the newly obtained cluster centres are recalculated in an iterative approach
until no data point is reassigned. The respective cluster group that users are assigned
represent their cognitive and emotional processing characteristics. For example, all
users that are grouped in the “Verbal” cluster are considered to have a verbal type of
cognitive style.

5.3.1.6 User Model

The above process results in constructing the final structure of the user model. More
specifically, the user model um of a user ui (um(ui)) is composed of demographics
(category = d), cognitive characteristics (category = cc) and emotional characteris-
tics (category = em) and contains triplets of the form (ct, ch, val), where ct represents
an information category (e.g., demographics, cognitive characteristics, emotional
characteristics, etc.), ch represents a characteristic (e.g., age, gender, verbal/imager
cognitive style, wholist/analyst cognitive style, working memory, general anxiety,
etc.) and val the value for the specific characteristic. For example, a user ui may have
the following user model:


um u d age cc v i verbal em anx highi( ) = ( ) ( ) ( ){ }, , , , , ,25 , / ,

indicating that ui has an age = 25 in the demographics (d) information category, he
is verbal (v/i) in the cognitive characteristics (cc) category and he is in general
highly anxious (anx) in the emotional characteristics (em) category.

5.3.2 Module 2: Personalization

Upon user classification, the Personalization module adapts semantically enriched
content at run-time on the client’s side. To accomplish this, the Personalization
module utilizes: (i) The cognitive and emotional characteristics inside the

5.3 Conceptual Design of mapU

Page 357

335

research has shown cross-cultural differences in fi eld dependence-independence
(Western vs. Eastern societies, African American vs. South African), and therefore
future work could investigate the effects of inter-cultural differences on human-
centered personalization interactions across different countries and continents
whose impact could affect a large number of individuals from different cultures.
Furthermore, the proposed approach could also have strong implications on elderly
people whose cognitive processing characteristics are limited and decline over time.
In this context, future research scenarios include conducting further user studies
with other samples and specialized profi les (including characteristics of older
adults) with the aim to strengthen the validity of the reported results and enrich our
understanding about the effects of the related users’ cognitive processing factors on
preference and performance during interaction.

Another timely future research direction would be to expand and further investi-
gate other intrinsic human factors (e.g., personality) with respect also to new tech-
nologies (e.g., device types) and interactive systems. For example, increase our
understanding about the effects of human factors on different interaction methods
driven by the utilization of external devices such as advanced eye-tracking systems
and wearable technologies and devices (e.g., wireless and body sensors, activity
trackers, smart watches, etc.). Currently, being part of the network of physical
objects or “things”, we are certainly coming closer to the users and we are able to
capture and correlate more data with respect to their activities and physiological
responses to given stimuli or situations. For instance, we could identify correlations
between specifi c cognitive mechanisms and generic somatic symptoms of anxiety
triggered during an activity step, and as a result to regulate this relationship to the
benefi t of the end-users by providing personalization solutions that will refl ect their
current state/feelings; allowing them to continue dealing with their tasks without
experiencing any negative emotions (e.g., worry, tension, nervousness, etc.).

In a parallel direction, in this era of big data produced from various and diverse
sources, an open question that still remains is how to generate human-centred adap-
tive visualizations that can increase users’ understanding and enable decision mak-
ing during the interaction process with multi-purpose data representations (e.g., in
the healthcare sector where a physician needs to make instantly a critical decision
for a patient by looking at different formats of visualizations presented in most
cases at different levels of detail). The requirement here is to identify potential cor-
relations of cognitive factors referring to high-level information processes as well as
elementary cognitive processes with different kinds (in terms of type and complex-
ity) of data visualizations (e.g. network diagrams, area and radar graphs, bar and
line charts, etc.), and consequently to suggest interventions that could increase the
usability and satisfaction of users based on their role and/or levels of expertise.

To sum up, it is true that there are many opportunities ahead but also numerous
constraints still in place, either engraved to the multi-dimensional character and
complexity of these research attempts or induced by the technological limitations of
today’s advancements that in many cases might hinder the practical feasibility of
solutions. However, we are convinced that the key to a viable co-existence of the

Epilogue

Page 358

336

human and the computer in today’s technological ecosystem that will offer usable
lasting interactions is to bring the human into the “centre” and design from and
with him. The driving factor in this pursuit is the end-user who should always have
the “benefi t of doubt” in any requirements collection, analysis, or compromization that
will eventually embody the design of an interface, service, system and/or application.
Thus, we strongly believe that an effective “recipe” should always acknowledge the
triptych from theory to principles and practice , and not vice versa.

Epilogue

Similer Documents