Flexible, autonome Verhaltenskontrolle: Erlernen und Adaptation von Sensormotorischen Raumrepräsentationen
Keynote by Martin Butz (COBOSLAB, University of Würzburg) at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: FGWM
Evaluating Tools of the X-COSIM Semantic Desktop
Presentation by Thomas Franz at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: FGWM
Hypothesis: PIM benefits from information linkage and information reuse across PIM application.
Method: Using RDF and Semantic Web technologies.
X-COSIM provides the X-COSIMO ontology:
Among others, X-COSIMO defines contextual information in a formal representation for context, the contextual ontology includes concepts such as email, attachment, sender, recipients. Attachments e.g. contextualizes information object, while sender and recipient contextualizes agents in the systems.
Evaluation:
Does an X-COSIM enabled desktop provide better support for PIM tasks than a conventional one?
Better means: increased effectiveness (absolute time spend on each task), increased efficiency (goals reached: distance of mouse movements, number of window switches, …), increased satisfaction (questionnaire, ratings, interview).
Results: Semantic desktop can improve PIM.
Implementation: Runs on KDE with Thunderbird. download now
Presentation by Andreas Hotho at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: KDML
The social bookmarking system BibSonomy has to deal with a lot of spam; which hamper the quality of search results and navigation. This talk focuses on detecting users as spammer, making all their posts invisible in the system. This decision is based on their tagging and personal data such as eMail etc. The authors present a framework that allows for automatic classification of spammers.
How to detect Spammers: Checking all their tags and, possible, the bookmarked sites. Spam posts are identified if:
Problems:
Features:
Classification algorithms: SVM (best), J48, Logistic regression (worse), and Naive Bayes
Presentation by Victor de Boer (Human-Computer Studies Laboratory; University of Amsterdam) at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: ABIS
The authors present a tool (SiteGuide) that supports amateur web-users to design web pages: Helping them to select and organize information for their pages. User input a set of example sites that are similar to their intended website. SiteGuide scrapes and analyses the sites and captures their commonalities in a web site model. From this, SiteGuide generates a site for the users. In addition, the system can provide the difference of a draft site of the user to its internal model of the example sites.
Presentation by Mirjam Köck at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: ABIS
Motivation: eLearning has become very popular, but still simple approaches of implementing adaptivity dominate. Computational intelligence is under-represented. The authors focus on challenges such as the autonomous knowledge acquisition, autonomous pattern identification at run-time, and expression of patterns in rules.
Adaptive learning guidance includes
Approach: Using communication and collaboration activities (rather than contents/ tags) are used as input for the user models. Identify groups of learners based on communication and learning behaviours (observing the style of learning, the level of activities, activity pattern).
Collecting information such as the user’s online time, actions related to communication (read, write, update, delete), user’s current knowledge, learning activities (time needed for test, time spent on content before taking tests, performance of tests), content of communication items
Relations of interest: How is a user’s time spent on communication related to the learning curriculum?; does the knowledge state influence communication?; what is the degree of similarity between a user’s activity level in the communication area and content area?; …
Promising Technologies: Artificial Neural Networks (can discover activity clusters, can adapt components e.g. change weights, do not depend on continuous human intervention; but: blackbox-syndrome, missing explanation capability, rule extraction is difficult); Combined Neuro-Fuzzy Approaches, Bayesian Networks (combination of domain knowledge and data; derivation of causal relationships) -> Combination allows using Neural Networks e.g. for learning and making the hidden sector of Neural Networks more visible.
Prospect: Improve adaptation; reducing human efforts to ensure quality and up-to-dateness of model data; semi-automatic pattern recognition, classification and evaluation at run-time; predication of behaviour based on correlations; integration of CI approaches into popular learning environment (Sakai; see also Stephan Weibelzahl)
Presentation by Stephan Weibelzahl (National College of Irland) at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: ABIS.
Stephan Weibelzahl developed HTML-Tutor, an interactive learning environment which offers an introduction to HTML and publishing on the Web.
Most system in academia remain at the prototype level with poor usability. This is acceptable as prototypes are vehicles to provide proof of concepts of an approach. However, in the long run scientists need to demonstrate the effects and impact of adaptive systems. Consequently, we need to start taking usability criteria into account when developing and evaluating scientific software. The authors aim at analysing the effects of usability on adaptation.
An adaptive peer finder is presented. The system is based on the adaptive learning system AHA! by de Bra and Calvi.
Further Readings:
Presentation by Sirko Schindler (University of Jena) at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: ABIS (collaboration with Andreas Nauerz). (see paper)
Treemaps were proposed by Johnson & Schneiderman in 1991. The author propose adaptive treemaps - displaying different treemaps to different users - to improve the navigation in web portals. To test several algorithms, the author developed a prototype, which is embedded into the IBM WebSphere Portal.
Presentation by Arne Koesling at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: ABIS
Focus: Introduce Adaptability into regular Learning Management Systems (LMS) - introduce sophisticated rule systems (policies) that allow all stakeholders (learners , teachers, system admins) to adapt the system for their needs.
Policies are statements that define the behaviour of the system and are intended to guide decisions and actions. There are also known as business rules. For example: “New customers have to pay in advance, regular customers are allowed to pay after delivery” or “point all users to the basic material, if they enter the course the first time and notify all tutors”. Policies are dynamic, declarative, reusable, have a well-defined semantics, and allow for reasoning.
The author chose Protune as policy language, which additionally supports negotiations, explanations and integration with external sources (see concrete examples in Protune on the w3.org website).
Presentation by Andreas Nauerz (IBM, University of Jena) at Lernen, Wissen, Adaptivität (LWA 2008), University of Würzburg, 6.-8. October 2008. Track: ABIS (see paper)
Focus: Improving accessibility of web portals. Company web portals (Enterprise Information Portals) often include immense corpora of contents, which are hardly ever used by the user as selection and navigation is often too tedious. Main concepts of the authors have been integrated in the IBM WebSphere Portal.
The author propose a complex user and context model (date, time, location). The user model reflect the user’s interest and preference: Information from static profiles (native language, home country, working location, age, …), the user’s interaction behaviour (pages and portlets they work with, tags users apply to resources); and the user’s social networks are used to derive knowledge on the user’s needs. For example, the user model includes information on the tagging, rating, and commenting behaviour of users: Tagging and rating behaviour are analysed to understand interest and preference of single users and entire communities. The static data is entered by the user, while more dynamic data is extracted from the user interaction using web usage mining.
Based on the user models two main services are provided:
Three independent context profile are created: (1) travelling, (2) office, (3) at home: User activities during are only stored in the respective context. For example, activity during travelling do not influence the user modelling for office work or at home.
Tags can be associate to web pages, documents, fragments of pages (very granular). They can be typed by the users or their semantics can be automatically extracted by calling respective back-end services.