Chapter 19 Systems That Adapt to Their Users
User-adaptive systems interactive system that adapts its behavior to individual users on the basis of processes of user model acquisition
User-adaptive systems User model acquisition involve some form of learning, inference, or decision making (distinguishes user-adaptive systems from adaptable systems) User-adaptive systems examples: adaptive UIs recommendation systems personalization
SUPPORTING SYSTEM USE
Adaptively offering help In cases where it is not sufficiently obvious to users how they should operate a given application, a help system can adaptively offer information and advice about how to use it When the user need advice? What commands the user is familiar with?
Adaptively offering help 1980’ Unix commands 1997 Office assistant 2010 CommunityCommands
Taking Over Parts of Routine Tasks Take over routine tasks that are simple but may place heavy demands on a user’s time System learns the patterns in frequent tasks by silently observe the user.
Taking Over Parts of Routine Tasks TaskTracer: the user-adaptive system learns which resources are associated to certain project and recognises which project the user works on currently Meeting appointment scheduling Categorisation of e-mails Smart reply
Taking Over Parts of Routine Tasks precision vs effort to save Careful user control in the beginning then (after learning) decreasing control. User learns what the system will be able to do successfully
Adapting the Interface
Adapting the Interface
Adapting the interface to individual abilities People with medical disorder AND temporary environmental factors: our fingers are slower in low temperature ambient noise will affect hearing ability illuminations impact reading speed pointing skills are much worse during walking
Adapting the interface to individual abilities Walking UI: standing and walking Uis similar but bigger UI items during walking Difficult to determine when to switch between states Switching during active usage is anoying
SUPPORTING INFORMATION ACQUISITION
Helping Users to Find Information Support for query-based search Support for browsing Spontaneous provison of information
Recommending Products recommend products that might be in the interest of the user based on the user’s history along with search (not instead) explanation increases trust
Recommending Products Collaborative filtering and content-based Critique-based recommender systems:
Tailoring Information Presentation Medical information can be presented in different ways to patients and doctors (interest, ability to understand) Learning user preferences for information presentation Color-blind users might require different color palette (adapting for individuals color perception)
Bringing people together = „recommending people ” finding friends expert search recommending social groups Internal social networking sites IBM SocialBlue: network-based and interest-based recommendation
Supporting Learning Adaptive e-learning courses Within-problem and outer-loop recommendations Stoichiometry Tutor (2011): give hints when the user makes a mistake (or asks for help) behaviour graph for each problem: acceptable paths to a solution along with possible incorrenct steps
OBTAINING INFORMATION ABOUT USERS
Explicit Self-Reports user supplies information to the system explicitly for the purpose of allowing the system to adapt objective properties of the user: age, profession, and place of residence (+) changing infrequently, (-) typing, (-) privacy concerns. Restrict requests for personal data to the few pieces of information that the system really requires! Explain the uses to which the data will be put!
Explicit Assessments Self-assessments of interests and knowledge like the level of the user’s interest in a particular topic, the level of his or her knowledge about it, or the importance that the user attaches to a particular evaluation criterion Responses to test items: tests of particular knowledge or skill. Outside of a learning context, users are likely to hesitate to invest time in tests of knowledge or skill, unless these can be presented in an enjoyable form…
Nonexplicit Input (event) log data: (+) silent, no effort from the user (-) difficult to utilise Social networks (like login with Facebook account) (+) useful explicit information already given (+) social relations also available Sensor data
Sensor inputs Sensor1 Sensor2 SensorN Environment SIGNAL PROCESSING Environment Current state of the user User model Application1 Application2 ApplicationN
Usability challenges Predictable and comprehensive Controllable Non-distractive Wide range of experience Privacy preserving
Predictability and Comprehensability The user is able to predict her actions’ influence on the system Users want to understand the general level of success of the system’s adaptation and they might want to understand why the system was (not) satisfactory in particular cases.
Controllability The user have to be able to approve or prevent particular automatic actions Especially actions having significant consequences Only recommendations or asking for approval
Distraction Do not reduce the users’ ability to concentrate on her primary task!
Filter bubble System supporting information acquisition [TED Talk] Solution: recommendation dictated not by the current user model
Privacy Security, privacy, trust Privacy: the ability of individuals to control the terms of under which their personal information is acquired and used (Culnan, 2000) social-based personalization, behavioral profiling, location-based personalization
User-adaptive systems Definition (user model) Functionality: Supporting system use Supporting information acquisition Obtaining information about users Usability challanges