Compiling Existing Strategies and Tools

From wiki.gpii
Jump to: navigation, search

Existing Approaches to Personalisation

This section presents existing approaches towards personalization related to accessibility and usability. This includes existing standards, projects as well as other related concepts and frameworks. Furthermore, machine-learning approaches for personalization are described. 

Personalization - Producing an end-user experience that is uniquely appropriate for each individual [sea07]

Personalization can be viewed from the question where it is performed:

  1. Personalization can be performed by the information system. In this case user data are collected implicitly by the system itself and inferences about the user need are made. In order to do any adaptation, information about the user is stored in a user profile, either on the local machine or in centralized database. A user profile consists of personal user information (such as name or ID) and information the system will use for adaptation.
  2. Personalization can be also performed by the user. That means modification of page layout, system changes explicitly performed by the user using the features provided by the system. For example alternative style, individual zoom setting or preferences regarding the content belong to this category. Advantage of this method is that it presents the users desire.

Both approaches to personalization could cause conflicts because computed user needs and user preference are not the same as real user needs and preferences. For example: the user might not understand the ramifications of a particular choice. For our approach to personalization it is important that we do not take control away from the user.

Furthermore, personalization can be viewed from the question what can be personalized. Different classifications have been published to which design areas interface adaptation can be applied. With the focus on web application we can distinguish between three areas of a system [bru02]: Selection of content, Presentation of information, Navigation. A classification based on adaptive email system distinguishes between 4 areas [pen07]: Content, Presentation, Media Selection, Adaptations of specific devices.  Adaptive user interfaces which aim at improving the accessibility will have to take into consideration the following design areas:

  1. adaptive content selection
  2. adaptive presentation
  3. adaptive navigation
  4. adaptive input mechanisms including the support of different input devices and technologies

Classification

Based on the following issues an analysis towards approaches of personalization related to accessibility and usability has been done:

  • Domain - The application domain of the approach.
  • Goal - The objective of the approach.
  • Adaptation - The design areas that are considered for personalization.
  • Method - Which personalization method is applied by an approach?
  • User Model - Which standard is used and which user groups are considered.
  • Context - Does the approach consider contextual information of the user?

1. Existing Applications (adaptive or adaptable)

Project
Domain
Goal
Adaptation



Method

User Model


Context



Content
Presentation
Navigation
Input
adaptive
adaptable
Standard
User Group
Kind of Personal Data (preferences, body functions, diagnostic)
MyDocStore
Accessible Documents
Transfering accessible personalized files between platforms.
x
x



x
unknown
user with print disabilities

unknown
MultiReader
Multimedia eBooks, Documents
On document for all
x
x
x


x
no
General, Blind, partially sighted, deaf readers, dyslexic readers

no
SNAPI
Public digital Terminals
Promote standardisation on smart media.

x
x
x

x
EN1332-4


no
APSIS4all
Public digital Terminals
Personalization of Public digital terminal.

x
x
x

x
EN1332-4


no
Web4All
Public Computer Workstations
Personalization of multi-user public access computer workstations

x
x
x

x
ISO 24751, IMS AccLIP


yes
TILE
e-learning
Provide a learning Object Repository (LOR) service
x
x
x


x
IMS AccLIP, IMS AccMD


no


2. Existing Frameworks towards personalized UIs

Project
Domain
Goal
Adaptation



Method

User Model


Context



Content
Presentation
Navigation
Input
adaptive
adaptable
Standard
User Group
Kind of Personal Data (preferences, body functions, diagnostic)
TransformAble
e-learning
Personalized online resources
x
x
x
x
 ?
x
ISO 24571 (AccessForAll)


unknown
GUIDE
TV set-top box applications
Adaptation of multimodal user interfaces
 ?
x
x
x
x

User Model
elderly users with mild impairments

yes
MYUI
ICT-products; (TV devices)
Mainstreaming of accessibility and individualized ICT
 ?
x
x
x
x

User Model
perceptual, cognitive, motor, general

yes

Standards related to Personalization

Here is a list of standards for personalization, in chronological order. See also File:UserDeviceProfiles-Synopsis.xlsx which provides a summary of the properties in each of these standards.

Other Relevant ETSI Documents

  • ETSI EG 202 249 V1.1.1 (2003-08): “Universal Communications Identifier (UCI); Guidelines on the usability of UCI based systems”
  • ETSI EG 202 325 V1.1.1 (2005-10): “Human Factors (HF); User Profile Management”: this guide describes the concept and provides guidelines.
  • ETSI EG 202 421 V1.1.1 (2007-01): “Human Factors (HF); Multicultural and language aspects of multimedia communications”. This guide proposes ways to offer cultural variants of a service that are best match to the user's preferences and abilities. The scope section of the document also lists types of users with disabilities, e.g. (quoted from the guide):
    • someone who only has a limited vocabulary in their own language;
    • someone who lipreads;
    • a user of sign language;
    • a person using the Blissymbols system; (...)
  • ETSI TS 102 747 (2009-12): “Human Factors (HF); Personalization and User Profile Management; Architectural Framework” (produced by ETSI Specialist Taskforce (STF) 342: Personalization and User Profile Management Standardization). This specification defines an “architectural framework supporting the personalization and user profile management concepts described in EG 202 325”. The specification discusses issues related to network requirements, functions and procedures (for example, profile creation and profile synchronization), security issues and privacy issues. It does not discuss accessibility, user interface adaptation or user profile items related to these topics.
  • Draft ETSI ES 202 642 V0.0.28 (2010-04): “Human Factors (HF), eHealth; Personalization of eHealth systems”

Machine-Learning Approaches for Personalisation

TODO:
* Tim F. Paymans, Jasper Lindenberg, Mark Neerincx: 
  "Usability trade-offs for adaptive user interfaces: ease of use and learnability"

User Interface Customization - in general - is a typical machine learning problem if approached from a statistical point of view. A given incomplete sample (the user) has to be completed using statistical information mined from other complete or incomplete samples (the profiles of other users). Consider Amazon, for example. They don't know what books you want to buy, but which books you already bought. From this preference, they can try to fill in your vectors and find the book that fits most to your vector (the overview recommendations 'that might be something for you') or they can find books, which complete a vector where you have already similar entries in your vector (the article recommendations 'people who bought this book did also like').

The AI and Machine Learning algorithms in those fields don't differ from what we could use. One of the core fields in this regard are the "recommender systems" and "classification" algorithms.

Pat Langley: User Modeling in Adaptive Interfaces (1999)

Langley, Pat. “User Modeling in Adaptive Interfaces” Proc. 7th International Conference on User Modeling, Banff, Canada, 1999

The author argues that there is a need for personalisation in many areas of interactive software. HCI research should not only focus on the manner in which the computer interface presents information and choices to the user, but also on the content that the interface presents to the user. The paper provides a definition of adaptive user interfaces, discusses a few examples based on literature, and reports on three research projects.

Langley defines an adaptive user interface as “a software artifact that improves its ability to interact with a user by constructing a user model based on partial experience with that user.” This definition is based on similar ones for machine learning. Since several different learning methods produce similar results, the paper focuses on other aspects: the reformulation of the problem or task in a form that the induction methods can address, the features used to describe behaviour, the source of data about user preferences.

Next, the paper describes examples of adaptive interfaces. Syskill & Webert is an interface that recommends web pages on a given topic; users can mark pages as desirable or undesirable, and the system encodes each user model in terms of the probabilities that certain words will occur based on the user’s like or dislike of a document. The system uses the naïve Bayesian classifier to learn probabilities and to predict preferences. This approach is called content-based filtering. Ringo is a film recommendation system that first requires the user to rate a series of films, from which it constructs a simple profile. Ringo then finds users with similar profiles and recommends films that they liked. This approach is called social or collaborative filtering; unlike content-based approaches, it does not require descriptions of the objects being recommended. Some systems combine both approaches. Other systems need to go beyond selecting items from a set and need to create new knowledge. Hermens and Schlimmer (1994) developed a system for filling out repetitive vacation forms. Each new form was used as training data and invoked an induction algorithm to revise its existing rules. After one year, it reduced keystrokes by 87%.

Next, the paper discusses three research projects that explored new directions in the automated construction of user models: an Adaptive Place Advisor (for finding restaurants), an Adaptive Route Advisor and an Adaptive Scheduling Assistant. The author then discusses related paradigms: programming by demonstration, intelligent tutoring systems, learning apprentices, behavioural cloning, and approaches that do not rely on machine learning. Rich’s stereotypes (1979) are more akin to classification than to induction, but the tow approaches are not mutually exclusive. The author also places adaptive interfaces in the context of research on computational models of human cognition.

The final section of the paper discusses challenges that arise from carrying over lessons from the study of machine learning to work on adaptive user interfaces. These challenges relate to the recasting of the problem of user modelling in terms of a standard induction task, the encoding of user data and user models, the collection of training cases, the constraints of “online learning” (as opposed to offline induction), and the importance of “rapidly”, i.e. based on a small number of training cases.

Hurst et al: Automatically Detecting Pointing Performance (2008)

Hurst, A., Hudson, S.E., Mankoff, J., and Trewin, S. Automatically detecting pointing performance. Proc. of the 13th International Conference on Intelligent User Interfaces. ACM, New York, 2008.

Abstract: "Since not all persons interact with computer systems in the same way, computer systems should not interact with all individuals in the same way. This paper presents a significant step in automatically detecting characteristics of persons with a wide range of abilities based on observing their user input events. Three datasets are used to build learned statistical models on pointing data collected in a laboratory setting from individuals with varying ability to use computer pointing devices. (...) "

Gajos et al: Predictability and Accuracy in Adaptive User Interfaces (2008)

Krzysztof Gajos, Katherine Everitt, Desney Tan, Mary Czerwinski, Daniel Weld. "Predictability and Accuracy in Adaptive User Interfaces" In: CHI '08: Proceeding of the twenty-sixth annual SIGCHI conference on Human factors in computing systems, pages 1271-1274, New York, NY, USA, 2008. ACM. (Alternative link)

Abstract (emphasis added): "While proponents of adaptive user interfaces tout potential performance gains, critics argue that adaptation's unpredictability may disorient users, causing more harm than good. We present a study that examines the relative effects of predictability and accuracy on the usability of adaptive UIs. Our results show that increasing predictability and accuracy led to strongly improved satisfaction. Increasing accuracy also resulted in improved performance and higher utilization of the adaptive interface. Contrary to our expectations, improvement in accuracy had a stronger effect on performance, utilization and some satisfaction ratings than the improvement in predictability."

Gajos et al: Automatically generating personalized user interfaces with Supple

Gajos, K.Z., Weld, D.S., and Wobbrock, J.O. "Automatically generating personalized user interfaces with Supple". Artificial Intelligence 174, 12-13 (2010), 910–950. doi:10.1016/j.artint.2010.05.005

The Supple system makes adaptations to the user interface automatically or in collaboration with the user. The authors "formally define interface generation as an optimization problem and demonstrate that, despite a large solution space (of up to 1017 possible interfaces), the problem is computationally feasible. In fact, for a particular class of cost functions, Supple produces exact solutions in under a second for most cases, and in a little over a minute in the worst case encountered, thus enabling run-time generation of user interfaces."

"Supple is not intended to replace human user interface designers—instead, it offers alternative user interfaces for those people whose devices, tasks, preferences, and abilities are not sufficiently addressed by the hand-crafted designs. Indeed, the results of our study show that, compared to manufacturers’ defaults, interfaces automatically generated by Supple significantly improve speed, accuracy and satisfaction of people with motor impairments." (Quoted from the abstract.)

Elsewhere, Gajos points out: "One of its limitations, however, is that the resulting interfaces are unlikely to capture all the nuances of the applications’ semantics, and the approach is fundamentally limited by what types of abilities can be modeled."

References

[bru02] Brusilovsky, P. & Maybury, M.T. (2002): From Adaptive Hypermedia To The Adaptive Web, in: Communications of the ACM. Volume 45, Number 5 (2002)

[MyUI] MyUI D2-2: MyUI: Mainstreaming Accessibility through Synergistic User Modelling and Adaptability http://www.myui.eu/docs.html

[pen07] Peng, X., & Silver, D. L. (2007). Interface Adaptation Based on User Expectation. Proceedings of the 21st International Conference on Advanced Information Networking and Applications Workshops, S.264-269, May 21-23.

[sea07] A. Sears, J.A. Jacko, The Human-Computer Interaction Handbook: Fundamentals, Evolving Technologies, and Emerging Applications (Human Factors & Ergonomics)

Valencia et al, 2013: Xabier Valencia, Myriam Arrue, J. Eduardo Pérez and Julio Abascal: "User Individuality Management in Websites based on WAI-ARIA Annotations and Ontologies" W4A, Rio de Janeiro, May 2013.