T Lecture Course ss 2011 Outline(first 2 weeks of lecture) CSCW 2: User Modeling Personalization and Recommender Systems User modeling Terms and overview Overview St wolfgang Wmd, 04.05 2011 User profile management Howwwhere is the information about user stored? quire information about cplicit vs implicit method Outline Motivation Motivating examples mation overload User modeling What to buy, dick, listen to...? Recommender system User modeling and personalization is a relatively old Modeling with probabilites, Bayes networks Acquiring user profiles uccessful application in commercial sites acquire information about See following examples, e. g. Amazon Explicit vs implicit metho Increasingly in mobile scenarios Amazon Example(Product Page) Amazon Example Information about product Behavior of other customers
1 Technische Universität München CSCW 2: User Modeling, Personalization and Recommender Systems 1. Introduction to User Modeling Wolfgang Wörndl, 04.05.2011 Lecture Course SS 2011 Technische Universität München Outline (first 2 weeks of lecture) • Motivating examples • User modeling – Terms and overview – User modeling methods • Overview: Stereotypes, overlay model • Data mining example • Modeling with probabilites, Bayes networks • User profile management – How/where is the information about user stored? • Acquiring user profiles – How to acquire information about users? – Explicit vs. implicit methods Next week! Wolfgang Wörndl 04.05.2011 2 Technische Universität München Outline • Motivating examples • User modeling – Terms and overview – User modeling methods • Overview: Stereotypes, overlay model • Data mining example • Modeling with probabilites, Bayes networks • User profile management – How/where is the information about user stored? • Acquiring user profiles – How to acquire information about users? – Explicit vs. implicit methods Wolfgang Wörndl 04.05.2011 3 Technische Universität München Motivation • Increasing information overload – What to buy, click, listen to …? – Possible solution • Personalization and adaption of information access • Recommender systems • User modeling and personalization is a relatively old academic field – e.g. E. Rich: “User Modeling via Stereotypes”, Cognitive Science 3, 329-354 (1979) • Successful application in commercial sites – See following examples, e.g. Amazon • Increasingly in mobile scenarios Wolfgang Wörndl 04.05.2011 4 Technische Universität München Amazon Example (Product Page) Behavior of other customers 5 Technische Universität München Amazon Example Information about product (meta data) 6
T Amazon Example Amazon Example Content-based filt Tags(by other users Reviews and ratings by other users) 等转, Example Movielens Example Google User login Example Google Example Foursquare recently visited web sites x② Search history whats aps, the time of day, and so an
2 Technische Universität München Amazon Example Content-based filter Collaborative filter 7 Technische Universität München Amazon Example Tags (by other users) Reviews and ratings (by other users) 8 Technische Universität München Example Movielens Wolfgang Wörndl 04.05.2011 9 Technische Universität München Example Google Wolfgang Wörndl 04.05.2011 10 Advertisements User login Technische Universität München Example Google Wolfgang Wörndl 04.05.2011 11 Recently visited web sites Search history Technische Universität München Example Foursquare Wolfgang Wörndl 04.05.2011 12 (Source: http://blog.foursquare.com/2011/03/08/foursquare-3/) „The idea is pretty simple: tell us what you’re looking for and we’ll help you find something nearby. The suggestions are based on a little bit of everything – the places you’ve been, the places your friends have visited, your loyalty to your favorite places, the categories and types of places you gravitate towards, what’s popular with other users, the day of the week, places with great tips, the time of day, and so on
T T Summary Outline Personalization is important and popular application Various methods used, e. g. Amazon product page User modeling daptation is often done in background, not always obvious Terms and overview Collecting data also possible for- anonymous"users User profile management is the information about user stored? Mapping to(real) users poss ble through(tree)registration or sweepstake Terms and Definitions O) Terms and Definitions(lI User: e Interacts with system User can act in different roles- user pseudonyms, identities Assignment of data and actions of persons without revealing the E.g. posting in a bulletin board/discussion forum In oum atoms it use ity ntification of a person+ stored Anonymity resp. anonymous information access to entity se eager ee im ie the opte ef acheson No mapping between user data or actions User profile with systems or other persons epresentation of a user in an information system, information Terms and Definitions(Ill) User Modeling process Data about Process to create, maintain and update a user model User model User profile vs user model Is difference User model Prof e denotes the concrete information that is stored about th Adaptation effect (Hrusovsky Maybury 2002)
3 Technische Universität München Summary • Personalization is important and popular application area – Various methods used, e.g. Amazon product page – Adaptation is often done in background, not always obvious to user • For example advertisements based on visited web sites • Problems with the privacy of personal data – See Google & Foursquare examples – Collecting data also possible for „anonymous“ users • Cookies, … • Mapping to (real) users possible through (free) registration or sweepstakes Wolfgang Wörndl 04.05.2011 13 Technische Universität München Outline • Motivating examples • User modeling – Terms and overview – User modeling methods • Overview: Stereotypes, overlay model • Data mining example • Modeling with probabilites, Bayes networks • User profile management – How/where is the information about user stored? • Acquiring user profiles – How to acquire information about users? – Explicit vs. implicit methods Wolfgang Wörndl 04.05.2011 14 Technische Universität München Terms and Definitions (I) • User: ! – Interacts with system – Generally identification required – User can act in different roles " user pseudonyms, identities • Identity (of a user) – Sociological term, related to a userʻs social roles – In our context: identity = identification of a person + stored informations in user profile – Virtual identity: identity in online world • Identity management implies the options of a person to choose her/his identity or role in the interaction with systems or other persons Wolfgang Wörndl 04.05.2011 15 Technische Universität München Terms and Definitions (II) • Pseudonym – Identifier for a (virtual) identity – Assignment of data and actions of persons without revealing the real or civil identity of the user • E.g. posting in a bulletin board/discussion forum • Anonymity resp. anonymous information access – Using a new pseudonym for every access/interaction – No mapping between user data or actions • User profile – Representation of a user in an information system, information stored about user Wolfgang Wörndl 04.05.2011 16 Technische Universität München Terms and Definitions (III) • User model – Abstract data structure to characterize users • User modeling – Process to create, maintain and update a user model • User profile vs. user model – Terms with simiar meaning, often used synonymous, difference not so important – In our lecture • Profile denotes the concrete information that is stored about the user in a system • Model designates the abstract representation of users Wolfgang Wörndl 04.05.2011 17 Technische Universität München User Modeling Process Data about user (Brusilovsky & Maybury, 2002) User model Adaptation effect User modeling Adaptation System collects processes processes Wolfgang Wörndl 04.05.2011 18
T Adaptive vs Intelligent Three Dimensions of user models 1. Intelligent but not adaptive(no user model!) 1. What is being modeled?(nature) 2. How is this information represented? g. choosing the" skin" of your favorite media player ustomization), or simple location-based servic (structure) 3. Intelligent and adaptive Main focus of lecture! 3. How are the models constructed a maintained? User modeling methods Intelligent system Adaptive system What is Being Modeled? application, including the user and the application E4. ement paston, tme, ae.TUmn User Context How is this Information Represented? Cognitive context What is user doing Context in mobile scenarios Limitations of mobile devices regarding displays, network Lists, vectors or sets, e.g. set of user interests Current location and time is important See Foursquare example But also: Limited attention span while moving Depends on application scenario discussed in more detail later on in lecture Various acquisition methods mplicit vs. explicit (next week!)
4 Technische Universität München Adaptive vs. Intelligent 1. Intelligent but not adaptive (no user model!) – E.g. anonymous search engine result 2. Adaptive but not really intelligent – E.g. choosing the “skin” of your favorite media player (customization), or simple location-based service 3. Intelligent and adaptive – Main focus of lecture! Intelligent system Adaptive system 2 3 1 Wolfgang Wörndl 04.05.2011 19 Technische Universität München Three Dimensions of User Models 1. What is being modeled? (nature) – Content of profile, characterization of user 2. How is this information represented? (structure) – Represention in computer program 3. How are the models constructed and maintained? – User modeling methods Wolfgang Wörndl 04.05.2011 20 Technische Universität München What is Being Modeled? • Personal and demographic data – Name, age, gender, email address(es) etc. • Interests and preferences – Selection from (pre-defined) categories, ratings, options (e.g. preferred language) etc. • User knowledge and goals – E.g. already visited courses (in learning scenario), domain knowledge (on a scale) • Authentication data and policies – E.g. user names and passwords, certificates, privacy preferences (e.g. regarding cookies) • Observed behavioral data (logs) – Clicks on web pages, bought products etc. • Payment information – E.g. credit card data • Contact data – Address books, Buddy lists, Group memeberships • Device data – Used devices with properties • Emotional state • User context – E.g. current postition, time, temperature, ... Wolfgang Wörndl 04.05.2011 21 Technische Universität München User Context • Definition: „Context is any information that can be used to characterize the situation of entities (i.e. whether a person, place or subject) that are considered relevant to the interaction between a user and an application, including the user and the application themselves.” (Dey, Abowd & Salber, D. A Conceptual Framework and a Toolkit for Supporting the Rapid Prototyping of Context-Aware-Applications. Human-Computer Interaction, 16(2), 97–166, 2007) • Views on context (Brusilovsky & Millan, 2007) Wolfgang Wörndl 04.05.2011 22 Technische Universität München User Context • Two main categories of context (others are possible) – Physical context: location, time, data from sensors – Cognitive context: What is user doing right now? • Context in mobile scenarios – Limitations of mobile devices regarding displays, network bandwidth, input capabilities, … – Current location and time is important • See Foursquare example – But also: Limited attention span while moving Wolfgang Wörndl 04.05.2011 23 Technische Universität München How is this Information Represented? • Various options – Scalar model, e.g. age = 23 – Attribute value pairs, e.g. „language=DE“ – Heuristic and probabilistic, e.g. Bayes networks – Fuzzy intervals – Lists, vectors or sets, e.g. set of user interests – Rules and rule sets – Declarative languages – Combined approaches • Depends on application scenario – Some options will be discussed in more detail later on in lecture • Various acquisition methods – Implicit vs. explicit (next week!) Wolfgang Wörndl 04.05.2011 24
T T Expressive Power of Modeling Outline User modeling Terms and overview Eg. prerequistes for a c User profile management Eg. is-a, part-of, andog is the information about user stored? How User Modeling Process? Stereoty pes Question: How to process(raw) data about a user to obtain a user model? group of users · Various methods E.g. female persons with an age of 40-50, computer science Customization, stereotypes, overlay model, data mining, Bayes networks 1. et of triggers to determine the condition to activate a User configures preferences or selects among pre defined options User is Not really, intelligent 3. Adaptation is done according to this classification Goal Incremental adjustment of user model possible Grundy() Grundy(Il) tereotypes are organized in hierarchy Dialogue with system to obtain keywords about user a stereotype is a cluster of characteristics: (facet, value, rating) Facets have values from 5 to +5 High rating high degree of certaintly(0 to 1000) Example: SPORTS-PERSON contains (romance, -5, 500) 5
5 Technische Universität München Expressive Power of Modeling Languages • Domain model – Body of domain knowledge is decomposed into set of smaller knowledge units – Set of concepts, topics, etc. • Vector vs. domain models – Vector, i.e. set of simple user model attributes • No relationships – Simple relationships • E.g. prerequisites for a course – Categories • E.g. “is-a”, “part-of”, analogies – (Semantic Web) Ontologies • More complex relationships More power Wolfgang Wörndl 04.05.2011 25 Technische Universität München Outline • Motivating examples • User modeling – Terms and overview – User modeling methods • Overview: Stereotypes, overlay model • Data mining example • Modeling with probabilites, Bayes networks • User profile management – How/where is the information about user stored? • Acquiring user profiles – How to acquire information about users? – Explicit vs. implicit methods Wolfgang Wörndl 04.05.2011 26 Technische Universität München How User Modeling Process? • Question: How to process (raw) data about a user to obtain a user model? • Various methods – Customization, stereotypes, overlay model, data mining, Bayes networks • Simpliest method: customization – User configures preferences or selects among predefined options – Not really „intelligent“ Wolfgang Wörndl 04.05.2011 27 Technische Universität München Stereotypes • Stereotyp: set of properties to characterize a user or a group of users – E.g. „female persons with an age of 40-50“, „computer science students“, … • User modeling with stereotypes 1. Definition of several stereotypes 2. Set of triggers to determine the condition to activate a stereotype " User is assigned to a specific group 3. Adaptation is done according to this classification • Goal – Quick characterization of users – Incremental adjustment of user model possible Wolfgang Wörndl 04.05.2011 28 Technische Universität München Grundy (I) • (Historical!) Example for stereotypes (Elaine Rich, 1979) • Scenario: suggest books in library • Dialogue with system to obtain keywords about user • A stereotype is a cluster of characteristics: (facet, value, rating) • Facets have values from -5 to +5 • High rating " high degree of certaintly (0 to 1000) • Example: SPORTS-PERSON contains – (thrill, +5, 700) – (romance, -5, 500) Wolfgang Wörndl 04.05.2011 29 Technische Universität München Grundy (II) • Stereotypes are organized in hierarchy Wolfgang Wörndl 04.05.2011 30
T T Grundy(l) Grundy(v) Stereotypes are activated through triggers User model Rating(how appropriate trigger is, similar to facet rating) activated the stereotype · User model is High_ Education_ Trigger NonTV_ StereoType dated over time Triggers are instantiated when user enters certain Suggest books based on generalized stereotype in hierarchy is activated stereotypes and triggers Overlay Model Abstract Example Overlay user model Domain model Maps knowledge, interests etc of user in relation to, complete has kno part of the domain model a E-Learning example: show addition informat Can not model_new knowledge Simple Overlay Model Weighted Overlay Model Concept
6 Technische Universität München Grundy (III) • Stereotypes are activated through triggers • A trigger consists of – Name – Corresponding stereotype – Rating (how appropriate trigger is; similar to facet rating) • Example: – High_Education_Trigger " NonTV_StereoType • Triggers are instantiated when user enters certain keywords – Example: Name “John” - Man_trigger (instantiate) • If a stereotyp is activated, then also the more generalized stereotype in hierarchy is activated Wolfgang Wörndl 04.05.2011 31 Technische Universität München Grundy (IV) • User model: (facet, value, rating, justification) – justification: trigger that activated the stereotype • User model is updated over time • Adaptation – Suggest books based on facets in model • Disadvantage – Manual effort to create stereotypes and triggers Wolfgang Wörndl 04.05.2011 32 Technische Universität München Overlay Model • Overlay user model – Maps knowledge, interests etc. of user in relation to „complete“ domain model • Expert profile represents ideal knowledge • User profile denotes what part of the domain model a user is interested in resp. has knowledge of – Interpretation of user actions compared to domain expert • E-Learning example: show addition information to novice user e.g. definitions of unknown terms • Advantage – Rather simple • Disadvantage – Can not model „new“ knowledge Wolfgang Wörndl 04.05.2011 33 Technische Universität München Abstract Example • Domain model Concept 1 Concept 2 Concept 3 Concept 4 Concept 5 Concept N (Brusilovsky & Millan, 2007) Wolfgang Wörndl 04.05.2011 34 Technische Universität München Simple Overlay Model Concept 1 Concept 2 Concept 3 Concept 4 Concept 5 Concept N yes no no no yes yes Wolfgang Wörndl 04.05.2011 35 Technische Universität München Weighted Overlay Model Concept 1 Concept 2 Concept 3 Concept 4 Concept 5 Concept N 10 3 0 2 7 4 Wolfgang Wörndl 04.05.2011 36
T T Concrete Example Overlay Model of Interests Domain model for user interests 密 Conclusion Literature · Summa E Rich: "User Modeling via Stereotypes, Cognitive yord examples for personalization P Brusilovsky &M.T. Marbury:-From adaptive Terms and definitions hypermedia to the adaptive web", Comm. of the ACM 45(5),2002 Overview of (simple)user modeler Stereotypes, overlay model P Brusilovsky &E Millan: "User Models for Adaptive · Next week emedia and Adaptive Educational Systems Bayes network Profile management and acquisition []P Brusilovsky, A Kobsa, W. NejdI (eds ) The Adaptive Web, Springer-Verlag, Berlin/Heidelberg 200
7 Technische Universität München Concrete Example • Domain model for user interests Wolfgang Wörndl 04.05.2011 37 Technische Universität München Overlay Model of Interests • For each domain concept an overlay model stores estimated level of interests 0.2 0.1 0.7 0.7 0.0 0.0 Wolfgang Wörndl 04.05.2011 38 Technische Universität München Conclusion • Summary – Real world examples for personalization and recommender systems – Terms and definitions • Dimensions of a user model – Overview of (simple) user modeling methods • Stereotypes, overlay model • Next week – Bayes networks – Profile management and acquisition Wolfgang Wörndl 04.05.2011 39 Technische Universität München Literature • E. Rich: “User Modeling via Stereotypes”, Cognitive Science 3, 329-354 (1979) • P. Brusilovsky & M.T. Marbury: „From adaptive hypermedia to the adaptive web“, Comm. of the ACM 45(5), 2002 • P.Brusilovsky & E.Millan: “User Models for Adaptive Hypermedia and Adaptive Educational Systems”, Chapter 1 in [*] [*] P.Brusilovsky, A.Kobsa, W.Nejdl (eds.): The Adaptive Web, Springer-Verlag, Berlin/Heidelberg, 2007 Wolfgang Wörndl 04.05.2011 40