Models of Context Why? –So we talk about it, write about it, argue about it –So we can show it to the user –So the user can understand it… –…and change it

Our model Context as a dynamic process with historic dependencies Context is comprised of a series of context states, like scenes in a movie

We have developed: –An interactional model of context –A software implementation of that model –Web service interface for the software –Integrated prototype using sensor inputs Context awareness for MOBIlearn

Context Awareness Purpose: –Context enables appropriate action - in this case learning Process: Context = a dynamic and historical process… …constructed from context states… … which are constructed through interaction between actors, situations, objects and activities... etc

Context Model Context What’s going on over time Context Substate Elements from the Learner and Setting that are relevant to the current focus of learning and desired level of context awareness Context State Elements from the Learning and Setting at one particular point in time, space, or goal sequence

Mapping this on to metadata Content Setting Metadata Learner + = Learning Objects + Resources + Services = Context Substate Context awareness

Basic Operation 1.Context features acquired or input 2.Context substate constructed from context features 3.Unsuitable content excluded 4.Remaining content ranked using current context state 5.Rankings output to delivery subsystems

Architecture Context Awareness Subsystem Content Server Sensors User input User profile Content metadata Content recommendations XML Content Environment

Objectives Use a model of context to dynamically select content Implement tracking system to provide real-time user location to the context system Evaluate technical issues surrounding implementation Perform trials of the system in mock-up gallery

Current status at UoB Context Awareness Subsystem –Java implementation –Available as a web service –Reads metadata from available content –Provides recommendations User tracking –Ultrasound positioning system –Tracking device attached to learner’s iPaq Content delivery –Pushed delivery of simple XHTML content to viewer on iPaq

Two factors: –Where is the user? –How long have they been there? Content recommended based on painting (from position) and inferred level of interest (from time) –10s = low = short title –20s = medium = short description –30s = high = full text Context awareness

Software Architecture Current software architecture implemented as object-oriented framework in Java Context Features are definable in text form and can be set to respond to specific metadata tags Context Features can interact with each other to provide more complex behaviour Learning Objects exist as index pointers to actual learning objects

Software implementation Painting Level of interest Location data time on value La Primavera, low detail La Primavera, high detail Birth of Venus, low detail Birth of Venus, high detail +1: Painting La Primavera Painting #41 Location sensors UltrasoundGPS User input Location sensors Painting #41

Software implementation Painting Level of interest Location sensors Location data time on value La Primavera, low detail La Primavera, high detail Birth of Venus, low detail Birth of Venus, high detail +1: Painting La Primavera Painting #41 +1: Interest low <10 seconds

Software implementation Painting Level of interest Location sensors Location data time on value La Primavera, low detail La Primavera, high detail Birth of Venus, low detail Birth of Venus, high detail +1: Painting La Primavera Painting #41 +1: Interest high >10 seconds

Software implementation Painting Level of interest Location sensors Location data time on value La Primavera, low detail La Primavera, high detail Birth of Venus, low detail Birth of Venus, high detail Salience:1 User profile: Level of Interest value Salience:2 high low +1: Painting +1: Interest

Software implementation Painting Level of interest Location sensors Location data time on value La Primavera, low detail La Primavera, high detail Birth of Venus, low detail Birth of Venus, high detail User profile: Level of interest value User input: content selection Salience:2Salience:1 La Primavera +1: Painting +2: User +1: Interest high >10 seconds Birth of Venus +1: Interest

Basic Functionality Perform comparison between: Context metadata (learner + environment) and… Learning Object metadata Comparison allows us to make recommendations about what is most relevant for the current context

Trials to be run at Nottingham Castle Museum in September Testing underway in mock-up art gallery Technology: –Ultrasound positioning sensors –Wireless PDAs –Content & metadata server –Other MOBIlearn system services Collaborative services Multimedia streaming Soon to be installed: –RFID tags & readers for iPaqs Test set-up

Title.

User Trials Small scale user trials, December 2003 Using limited contextual data... –Location –Location of others –Current question –Questions answered by others –Time on question...to modify content: –Painting/artist details –Recommended next question

Title.

It works: –Learners able to quickly find relevant information and successfully answer the questions Open issues: –Interface: crucial to get the representation right –Understanding: some people weren’t quite sure why the system did what it did, and were surprised by the constantly changing list of options –Distraction vs Engagement: offering multiple choices led to sidetracking or encouraged people to further their exploration of the content –Mixed content: need to to distinguish questions, content, physical resources Issues from user trials

Issues from stakeholders Lack of intro & follow-up –Museum visitors are un-prepared, and do not get to follow-up their visits afterwards Lack of collaboration –Wandering around the exhibits is often a lonely business, with visitors separated by group membership, and time No way to determine & provide appropriate amount of content –Each visitor comes to the museum with a unique set of interests. Crucially, these may change during their visit No focused attention –Visitors often enter at one of the gallery and move straight to the exit at the other end, without stopping to pay proper attention to what’s in-between

The challenge How can we create an engaging experience? Track visitor locations, offer them content that is relevant –Track location and how long they’ve been there Track what they do over time, and recommend things according to their ‘path’ –This path can also be compared to others’ paths Offer and encourage collaboration, through communication services, collaborative activities –A memorable, social experience will be a lot more engaging Focus their attention – point out items of interest – stop them walking straight through the gallery –Each painting has a story, beyond what it depicts, eg some buildings may be more prominent because of who has paid for the painting (and the building)... Support exploratory learning and curiosity –Even random pointers are likely to be better than no pointers at all

Next steps Roll-out in Nottingham Castle Museum Trials to compare: –Different representations of context –Automated vs. Manual input of contextual data –Different methods of initiating and maintaining collaborative activities –Ways of guiding them to ask questions, not find answers How do we pique their curiosity, and keep them interested?

Navigation metaphor Context aware navigation of content is replacing the more familiar web browser metaphor User interface issues include: –Should we provide web-style navigation (eg Back, Forward, History) –Will users exploit the context metaphor for content navigation (eg movement = navigation) or will it hinder them?

Next 3-6 months User trials –Uffizi –Nottingham Castle Museum Development –Display of context model to user –Provision of user controls, eg ‘hold’ button and ‘Why was this recommended to me?’ –Exploration of ‘context navigation metaphor’ –Use of context history to influence current recommendations

System integration CA_CAS – fully implemented services (as of 26/08/04) –ContextDataInput Send contextual data to the system –EndSession End a user session –GetContextFeature Retrieve information from one context feature (according to ContextFeature schema) –GetContextualInformation Retrieve information from all context features (according to ContextObject schema) –GetRecommendationInfo Retrieve information about why a recommendation was made (plain text) –GetRecommendedOptions Retrieve current Learning Object recommendations (Recommendation schema) –InitializeContext Start a session for a user CA_CMS – not implemented –Main functionalities moved to CA_CAS (only CAS called these services)

Context Services Context Awareness Service: CAS –Context Object is the sum of currently available context information, as a nested set of context feature objects –This defines a subset of metadata to search for on available content Context Object Management Service: CMS –Supports creation, retrieval, and modification of context feature objects

Title