HCI Bibliography Home | HCI Conferences | UMAP Archive | Detailed Records | RefWorks | EndNote | Hide Abstracts
UMAP Tables of Contents: 00010203040506070809101112131415

Proceedings of User Modeling 2003 2003-06-22

Fullname:Proceedings of the 9th International Conference on User Modeling
Editors:Peter Brusilovsky; Albert Corbett; Fiorella de Rosis
Location:Johnstown, Pennsylvania
Dates:2003-Jun-22 to 2003-Jun-26
Publisher:Springer-Verlag
Series:Lecture Notes in Computer Science, 2003, Volume 2702
Standard No:ISBN: 978-3-540-40381-4 (Print) 978-3-540-44963-8 (Online); hcibib: UMAP03
Papers:66
Pages:434
Links:Online Proceedings | Conference Home Page
Summary:The International User Modeling Conferences are the events at which research foundations are being laid for the personalization of computer systems. In the last 15 years, the field of user modeling has produced significant new theories and methods to analyze and model computer users in short- and long-term interactions. A user model is an explicit representation of properties of individual users or user classes. It allows the system to adapt its performance to user needs and preferences. Methods for personalizing human-computer interaction based on user models have been successfully developed and applied in a number of domains, such as information filtering, adaptive natural language and hypermedia presentation, tutoring systems, ecommerce and medicine. There is also a growing recognition of the need to evaluate the results of new user modeling methods and prototypes in empirical studies and a growing focus on evaluation methods.
  1. Abstracts of Invited Talks
  2. Adaptive Hypermedia
  3. Adaptive Web
  4. Natural Language and Dialog
  5. Plan Recognition
  6. Evaluation
  7. Emerging Issues of User Modeling
  8. Group Modeling and Cooperation
  9. Applications
  10. Student Modeling Methods
  11. Learning Environments: Natural Language and Pedagogy
  12. Mobile and Ubiquitous Computing
  13. Doctoral Consortium Papers

Abstracts of Invited Talks

Adaptive Interfaces for Ubiquitous Web Access BIBAFull-Text 1
  Michael Pazzani
The World Web Wide gives unprecedented access to Newspapers, magazines, shopping catalogs, restaurant guides, and classified ads and other types of information. All this information, however, used to be accessible only while users are tethered to a computer at home or in an office. Wireless data and voice access to this vast store allows unprecedented access to information from any location at any time.
Computers That Recognize and Respond to User Emotion BIBAFull-Text 2
  Rosalind Picard
Did you like that or not? Did the system's choice of adaptation aggravate you more, or did it bring about an expression of gratefulness? Does this interest you or bore you? Recognition of the effects an action has on a user is a key part of adapting successfully to users; how can machines be enabled to recognize affective expressions such as frustration, interest, anger, or joy? And, what are guidelines for designing their response, especially given that recognition is likely to not be perfect? This talk will present new technologies under development for sensing and responding appropriately to human affective expressions. Current applications include usability feedback, health behavior change, learning companions, and human-robot interaction.
The Advantages of Explicitly Representing Problem Spaces BIBAFull-Text 3
  Kurt VanLehn
Newell and Simon (1972) coined the term "problem space" for a virtual structure: all possible lines of reasoning that can be employed by an agent to solve a problem. For certain toy problems (e.g., Tower of Hanoi), the problem space can be represented explicitly as labeled, directed graph. For non-toy problems, cognitive scientists have sometimes employed the concept of a problem space to analyze tasks, but seem to feel that explicit representation of the whole problem space for a problem is probably not worthwhile, and perhaps not even feasible. I will present techniques developed over a decade of research that make explicit representation of large problem spaces feasible. I will demonstrate how explicit representations of problem spaces have been used in systems that do non-trivial user modeling, task analysis, intelligent tutoring, and natural language dialogues.

Adaptive Hypermedia

The Three Layers of Adaptation Granularity BIBAFull-Text 4-14
  Alexandra Cristea; Licia Calvi
In spite of the interest in AHS, there are not as many applications as could be expected. We have previously pinpointed the problem to rely on the difficulty of AHS authoring. Adaptive features that have been successfully introduced and implemented until now are often too fine grained, and an author easily loses the overview. This paper introduces a three-layer model and classification method for adaptive techniques: direct adaptation rules, adaptation language and adaptation strategies. The benefits of this model are twofold: on one hand, the granulation level of authoring of adaptive hypermedia can be precisely established, and authors therefore can work at the most suitable level for them. On the other hand, this is a step towards standardization of adaptive techniques, especially by grouping them into a higher-level adaptation language or strategies. In this way, not only adaptive hypermedia authoring, but also adaptive techniques exchange between adaptive applications can be enabled.
Adaptive Presentation of Multimedia Interface Case Study: "Brain Story" Course BIBAFull-Text 15-24
  Halima Habieb-Mammar; Franck Tarpin-Bernard; Patrick Prévôt
The paper presents the development of a multimedia adaptive interface based on the cognitive profile of a user. It describes the cognitive profile and the document architecture that have been adopted. Indeed, the user profile is generated in a stage former to the adaptation, it is structured into a separate database. As for the document, it is presented in XML files. The paper describes in detail the way these components are combined; it presents the adaptive process. During this process, the combination of medias that best fits the cognitive profile of each user is selected. This technique is applied on a hypermedia course called "Saga du Cerveau" (The Brain Story).
Discovering Prediction Rules in AHA! Courses BIBAFull-Text 25-34
  Cristóbal Romero; Sebastián Ventura; Paul de Bra; Carlos de Castro
In this paper we are going to show how to discover interesting prediction rules from student usage information to improve adaptive web courses. We have used AHA! to make courses that adapt both the presentation and the navigation depending on the level of knowledge that each particular student has. We have performed several modifications in AHA! to specialize it and power it in the educational area. Our objective is to discover relations between all the picked-up usage data (reading times, difficulty levels and test results) from student executions and show the most interesting to the teacher so that he can carry out the appropriate modifications in the course to improve it.

Adaptive Web

Word Weighting Based on User's Browsing History BIBAFull-Text 35-44
  Yutaka Matsuo
We developed a word-weighting algorithm based on the information access history of a user. The information access history of a user is represented as a set of words, and is considered to be a user model. We weight words in a document according to their relevancy to the user model. The relevancy is measured by the biases of co-occurrence, called IRM (Interest Relevance Measure), between a word in a document and words in the user model. We evaluate IRM through a constructed browsing support system, which monitors word occurrences on the user's browsed Web pages and highlights keywords in the current page. Our system consists of three components: a proxy server that monitors access to the Web, a frequency server that stores the frequencies of words appearing on the accessed Web pages, and a keyword extraction module.
SNIF-ACT: A Model of Information Foraging on the World Wide Web BIBAFull-Text 45-54
  Peter Pirolli; Wai-Tat Fu
SNIF-ACT (Scent-based Navigation and Information Foraging in the ACT architecture) has been developed to simulate users as they perform unfamiliar information-seeking tasks on the World Wide Web (WWW). SNIF-ACT selects actions based on the measure of information scent, which is calculated by a spreading activation mechanism that captures the mutual relevance of the contents of a WWW page to the goal of the user. There are two main predictions of SNIF-ACT: (1) users working on unfamiliar tasks are expected to choose links that have high information scent, (2) users will leave a site when the information scent of the site diminishes below a certain threshold. SNIF-ACT produced good fits to data collected from four users working on two tasks each. The results suggest that the current content-based spreading activation SNIF-ACT model is able to generate useful predictions about complex user-WWW interactions.
Adapting to the User's Internet Search Strategy BIBAFull-Text 55-64
  Jean-David Ruvini
World Wide Web search engines typically return thousands of results to the users. To avoid users browsing through the whole list of results, search engines use ranking algorithms to order the list according to predefined criteria. In this paper, we present Toogle, a front-end to the Google search engine for both desktop browsers and mobile phones. For a given search query, Toogle first ranks results using Google's algorithm and, as the user browses through the result list, uses machine learning techniques to infer a model of her search goal and to adapt accordingly the order in which the results are presented. We describe preliminary experimental results that show the effectiveness of Toogle.
Learning a Model of a Web User's Interests BIBAFull-Text 65-75
  Tingshao Zhu; Russ Greiner; Gerald Häubl
There are many recommender systems that are designed to help users find relevant information on the web. To produce recommendations that are relevant to an individual user, many of these systems first attempt to learn a model of the user's browsing behavior. This paper presents a novel method for learning such a model from a set of annotated web logs -- i.e., web logs that are augmented with the user's assessment of whether each webpage is an information content (IC) page (i.e., contains the information required to complete her task). Our systems use this to learn what properties of a webpage, within a sequence, identify such IC-pages, and similarly what "browsing properties" characterize the words on such pages ("IC-words"). As these methods deal with properties of webpages (or of words), rather than specific URLs (words), they can be used anywhere throughout the web; i.e., they are not specific to a particular website, or a particular task. This paper also describes the enhanced browser, aie, that we designed and implemented for collecting these annotated web logs, and an empirical study we conducted to investigate the effectiveness of our approach. This empirical evidence shows that our approach, and our algorithms, work effectively.
Modelling Users' Interests and Needs for an Adaptive Online Information System BIBAFull-Text 76-80
  Enrique Alfonseca; Pilar Rodríguez
A system has been built that adapts the contents of existing web sites to the needs of the users. With this system, it is possible to define the user interests about any kind of topic, and they are used to filter the information from the site. To do it, we have used topic identification and document classification techniques. As an additional functionality, a summarisation system has been integrated with the system, and the user can specify which is the level of compression to be performed on the original hypertext pages from the static web site. In a preliminary evaluation the system was well received among potential users.
Declarative Specifications for Adaptive Hypermedia Based on a Semantic Web Approach BIBAKFull-Text 81-85
  Serge Garlatti; Sébastien Iksal
Adaptation/personalization is one of the main issues for web services. Adaptive web applications have the ability to deal with different users' needs for enhancing usability and comprehension and for dealing with large repositories. We propose an open-ended adaptive hypermedia environment which is based on the virtual document and semantic web approaches and which is able to manage adaptive techniques at knowledge level. In this paper, we have focused on the way to specify and to manage adaptation in this environment. We propose an approach which is based on a unique evaluation principle of links/contents per document and where the author may assign user stereotypes to adaptive techniques.
Keywords: Adaptive navigation and presentation; Virtual Document; Composition Engine; Semantic Web

Natural Language and Dialog

Emotional Dialogs with an Embodied Agent BIBAFull-Text 86-95
  Addolorata Cavalluzzi; Berardina De Carolis; Valeria Carofiglio; Giuseppe Grassano
We discuss how simulating emotional dialogs with an Embodied Agent requires endowing it with ability to manifest appropriately emotions but also to exploit them in controlling behavior. We then describe a domain-independent testbed to simulate dialogs in affective domains and verify how they change when the context in which interaction occurs is varied. Emotion activation is simulated by dynamic belief networks while dialog simulation is implemented within a logical framework.
Evaluating a Model to Disambiguate Natural Language Parses on the Basis of User Language Proficiency BIBAFull-Text 96-105
  Lisa N. Michaud; Kathleen F. McCoy
This paper discusses the evaluation of an implemented user model in ICICLE, an instruction system for users writing in a second language. We show that in the task of disambiguating natural language parses, a blended model combining overlay techniques with user stereotyping representing typical linguistic acquisition sequences successfully captures user individuality while supplementing incomplete information with stereotypic reasoning.
Incorporating a User Model into an Information Theoretic Framework for Argument Interpretation BIBAFull-Text 106-116
  Ingrid Zukerman; Sarah George; Mark George
We describe an argument-interpretation mechanism based on the Minimum Message Length Principle [1], and investigate the incorporation of a model of the user's beliefs into this mechanism. Our system receives as input an argument entered through a web interface, and produces an interpretation in terms of its underlying knowledge representation -- a Bayesian network. This interpretation may differ from the user's argument in its structure and in its beliefs in the argument propositions. The results of our evaluation are encouraging, with the system generally producing plausible interpretations of users' arguments.
Using Dialogue Games to Maintain Diagnostic Interactions BIBAFull-Text 117-121
  Vania Dimitrova
This paper presents an approach to dynamically extract individual user models by engaging users in diagnostic interactions. A framework for maintaining diagnostic dialogues based on approaches known as dialogue games is outlined and illustrated in STyLE-OLM -- an interactive student modelling system. The framework is validated in an evaluative study of STyLE-OLM, and potential improvements are sketched out.

Plan Recognition

Extending Plan Inference Techniques to Recognize Intentions in Information Graphics BIBAFull-Text 122-132
  Stephanie Elzer; Nancy Green; Sandra Carberry; Kathleen McCoy
Plan inference techniques have been used extensively to understand natural language dialogue. But as noted by Clark[5], language and communication are more than just utterances. This paper presents the problems that we have had to address and the solutions that we have devised in designing a system to recognize intentions from information graphics. Our work is part of a larger project to develop an interactive natural language system that provides an alternative means for individuals with sight-impairments to access the content of information graphics.
Leveraging Collaborative Effort to Infer Intent BIBAFull-Text 133-137
  Joshua Introne; Richard Alterman
We describe method for intent inference in collaborative systems, and an application that makes use of intent inference to facilitate coordination. Our approach to intent inference is unique in that we derive intent by piggybacking on coordination specific communication that occurs in collaboration. We have developed an interface component that uses the output of our inference engine to support users' awareness of eachother's activities and offload some of the individual user's work.
Plan Recognition to Aid the Visually Impaired BIBAFull-Text 138-142
  Marcus J. Huber; Richard Simpson
Less than half of the individuals of working age with visual impairments are employed and a significant barrier to employment is effective computer access. Screen reader applications offer some help but have limited context sensitivity and are of limited use in applications with dynamic "interfaces" like web pages. Sophisticated screen readers provide aid through application-specific scripts but their full potential is reduced by limited awareness of the scripts and the difficulty in programming and modifying scripts. Technologies such as plan recognition and automated script generation and optimization that provide a more adaptive interface for the user will significantly improve computer accessibility to the visually impaired. In this paper, we discuss the addition of probabilistic plan recognition capabilities and supporting framework to a leading screen reader in order to improve accessibility of computers to the visually impaired at work and at home.

Evaluation

Performance Evaluation of User Modeling Servers under Real-World Workload Conditions BIBAFull-Text 143-153
  Alfred Kobsa; Josef Fink
Before user modeling servers can be deployed to real-world application environments with potentially millions of users, their runtime behavior must be experimentally verified under realistic workload conditions to ascertain their satisfactory performance in the target domain. This paper discusses performance experiments which systematically vary the number of profiles available in the user modeling server, and the frequency of page requests that simulated users submit to a hypothetical personalized website. The parameters of this simulation are based on empirical web usage research. For small to medium sized test scenarios, the processing time for a representative mix of user modeling operations was found to only degressively increase with the frequency of page requests. The distribution of the user modeling server across a network of computers additionally accelerated those operations that are amenable to parallel execution. A large-scale test with several million active user profiles and a page request rate that is representative of major websites confirmed that the user modeling performance of our server will not impose a significant overhead for a personalized website. It also corroborated our earlier finding that directories provide a superior foundation for user modeling servers than traditionally used data bases and knowledge bases.
Evaluating the Inference Mechanism of Adaptive Learning Systems BIBAFull-Text 154-162
  Stephan Weibelzahl; Gerhard Weber
The evaluation of user modeling systems is an important though often neglected area. Evaluating the inference of user properties can help to identify failures in the user model. In this paper we propose two methods to assess the accuracy of the user model. The assumptions about the user might either be compared to an external test, or might be used to predict the users' behavior. Two studies with five adaptive learning courses demonstrate the usefulness of the approach.
The Continuous Empirical Evaluation Approach: Evaluating Adaptive Web-Based Courses BIBAFull-Text 163-167
  Alvaro Ortigosa; Rosa M. Carro
In this paper we present the continuous empirical evaluation approach, whose goal is to improve the quality of adaptive web-based courses. The adaptive-course description, along with the users features and interactions with the courses, are analyzed in order to detect concrete possible fails or lacks and to propose specific solutions and actions to be performed to improve these courses. The way it is used to evaluate existing adaptive courses is also presented.

Emerging Issues of User Modeling

Privacy Preservation Improvement by Learning Optimal Profile Generation Rate BIBAFull-Text 168-177
  Tsvi Kuflik; Bracha Shapira; Yuval Elovici; Adlai Maschiach
PRAW, a privacy model proposed recently, is aimed at protecting Web surfers' privacy by hiding their interests, i.e., their profiles. PRAW generates several faked transactions for each real user's transaction. The faked transactions relate to various fields of interest in order to confuse eavesdroppers attempting to derive users' profiles. They provide eavesdroppers with inconsistent data for the profile generation task. PRAW creates two profiles, a real user profile and a faked one aimed at confusing eavesdroppers. In this paper we demonstrate that the number of user transactions used for user profile generation significantly affects PRAW's ability to hide users' interests. We claim that there exists an optimal profile update rate for every user according to his surfing behavior. A system implementing PRAW needs to learn, for each specific user, the user's behavior, and dynamically adjust the optimal number of transactions that should be used to generate the user profile.
Interfaces for Eliciting New User Preferences in Recommender Systems BIBAFull-Text 178-187
  Sean M. McNee; Shyong K. Lam; Joseph A. Konstan; John Riedl
Recommender systems build user models to help users find the items they will find most interesting from among many available items. One way to build such a model is to ask the user to rate a selection of items. The choice of items selected affects the quality of the user model generated. In this paper, we explore the effects of letting the user participate in choosing the items that are used to develop the model. We compared three interfaces to elicit information from new users: having the system choose items for users to rate, asking the users to choose items themselves, and a mixed-initiative interface that combines the other two methods. We found that the two pure interfaces both produced accurate user models, but that directly asking users for items to rate increases user loyalty in the system. Ironically, this increased loyalty comes despite a lengthier signup process. The mixed-initiative interface is not a reasonable compromise as it created less accurate user models with no increase in loyalty.
Modeling Multitasking Users BIBAFull-Text 188-197
  Malcolm Slaney; Jayashree Subrahmonia; Paul Maglio
This paper describes an algorithm to cluster and segment sequences of low-level user actions into sequences of distinct high-level user tasks. The algorithm uses text contained in interface windows as evidence of the state of user-computer interaction. Window text is summarized using latent semantic indexing (LSI). Hierarchical models are built using expectation-maximization to represent users as macro models. User actions for each task are modeled with a micro model based on a Gaussian mixture model to represent the LSI space. The algorithm's performance is demonstrated in a test of web-browsing behavior, which also demonstrates the value of the temporal constraint provided by the macro model.
VlUM, a Web-Based Visualisation of Large User Models BIBAFull-Text 198-202
  James Uther; Judy Kay
This paper describes VlUM, a new tool for visualising large user models. It is intended to help users gain both an overview of the system's model of a user as well as the ability to find interesting parts of the model. In particular, it is intended to enable users to quickly identify outlier or interesting parts of the model.
A Multiagent Approach to Obtain Open and Flexible User Models in Adaptive Learning Communities BIBAFull-Text 203-207
  Felix Hernandez; Elena Gaudioso; Jesus G. Boticario
Nowadays, many user-modeling systems are applied to web-based adaptive systems. The large number of very different users using these systems make user model construction difficult. The solution is to use machine learning techniques that dynamically update the models by monitoring user behavior. However, the design of machine learning tasks for user modeling is static. This poses a problem in adaptive learning environments based on virtual communities. Each virtual community has its own administrators, and each administrator may prefer to include some more information on the user model. Another problem in the application of machine learning techniques for user model construction is the need to retrain the machine learning algorithms when new user interaction data become available. To face these problems, in this paper we present a multiagent adaptive module set in an adaptive learning collaborative environment. Our goal is two fold: (i) we want each administrator to be able to define new machine learning attributes in the user model (ii) we want to provide a mechanism to dynamically retrain the algorithms.
A Model for Integrating an Adaptive Information Filter Utilizing Biosensor Data to Assess Cognitive Load BIBAKFull-Text 208-212
  Curtis S. Ikehara; David N. Chin; Martha E. Crosby
Information filtering is an effective tool for improving performance but requires real-time information about the user's changing cognitive states to determine the optimal amount of filtering for each individual at any given time. Current research at the Adaptive Multimodal Interactive Laboratory assesses the user's cognitive ability and cognitive load from physiological measures including: eye tracking, heart rate, skin temperature, electrodermal activity, and the pressures applied to a computer mouse during task performance. A model of adaptive information filtering is proposed that would improve learning and task performance by optimizing the human-computer interface based on real-time information of the user's cognitive state obtained from these passive physiological measures.
Keywords: Physiological sensor; biosensor; information filter; cognitive load
Ontology-Based User Modeling for Knowledge Management Systems BIBAFull-Text 213-217
  Liana Razmerita; Albert Angehrn; Alexander Maedche
This paper is presenting a generic ontology-based user modeling architecture, (OntobUM), applied in the context of a Knowledge Management System (KMS). Due to their powerful knowledge representation formalism and associated inference mechanisms, ontology-based systems are emerging as a natural choice for the next generation of KMSs operating in organizational, interorganizational as well as community contexts. User models, often addressed as user profiles, have been included in KMSs mainly as simple ways of capturing the user preferences and/or competencies. We extend this view by including other characteristics of the users relevant in the KM context and we explain the reason for doing this. The proposed user modeling system relies on a user ontology, using Semantic Web technologies, based on the IMS LIP specifications, and it is integrated in an ontology-based KMS called Ontologging. We are presenting a generic framework for implicit and explicit ontology-based user modeling.

Group Modeling and Cooperation

Motivating Cooperation on Peer to Peer Networks BIBAFull-Text 218-227
  Helen Bretzke; Julita Vassileva
This paper addresses the problem of free riding on peer-to-peer resource-sharing networks and explores methods for motivating more cooperative user behaviour via an adaptive interface. The paper argues that the free-riding problem is not so much an economic issue as a socio-psychological one due to a paradigm shift the user community is undergoing. Users do not yet understand that they, and all of their peers, are both clients and servers and must therefore be taught new behaviour. Our method stimulates community awareness and highlights the cause and effect relationship between user behaviour and performance (QoS) consequences. Modeling the user's interests, attitude and relationships with other users enables the interface to adapt to the individual's cooperativeness bias and give feedback on current community structure and activity. Feedback is delivered in the form of graphs, animations and informative text.
Discourse Analysis Techniques for Modeling Group Interaction BIBAFull-Text 228-237
  Alexander Feinman; Richard Alterman
This paper presents discourse analysis techniques that model the interaction of a small group of users engaged in same-place/different-time interaction. We analyzed data from VesselWorld, our experimental testbed, and formulated a modeling technique based on the recurrence of coordination problems and the structure that users create to handle these problems. Subsequent experiments revealed that our original analysis had failed to capture issues with the cognitive load required to maintain common ground. By tracking references users make to both domain and conversational objects, we were able to extract patterns of information access and model the cognitive load incurred to maintain common ground. The improved model of user interaction was successful in explaining systems designed to support interaction.
Group Decision Making through Mediated Discussions BIBAFull-Text 238-247
  Daniel Kudenko; Mathias Bauer; Dietmar Dengler
To date, product recommendation systems have mainly been looked at from a single-agent perspective, where only the interests of a single user are taken into account. We extend this scenario and consider the case where multiple users are planning a joint purchase, and therefore many (potentially conflicting) interests have to be considered.
   In this paper we present an overview of a system that assists a group of users to reach a joint decision on an online catalogue purchase. This is done by acquiring individual user models and using these models to mediate a kind of group discussion with the goal to arrive at a compromise that is acceptable to all group members.
Modeling Task-Oriented Discussion Groups BIBAFull-Text 248-257
  Roy Wilson
Several recent studies present complementary mathematical models of actor behavior in small, task-oriented, groups. This paper describes both models, which share the Markov property, discusses their strengths and limitations, and suggests that user modeling researchers and small group process researchers might benefit from collaboration. Several possibilities for model-based collaboration are suggested in connection with Computer Supported Collaborative Learning.
Modeling the Multiple People That Are Me BIBAFull-Text 258-262
  Judith Masthoff
A new approach is outlined in which group modeling techniques are used to model an individual user. This helps to reduce cold-start problems, and allows aggregating multiple criteria.

Applications

Iems: Helping Users Manage Email BIBAFull-Text 263-272
  Eric McCreath; Judy Kay
This paper reports our work to build an email interface which can learn how to predict a user's email classifications at the same time as ensuring user control over the process. We report our exploration to answer the question: does the classifier work well enough to be effective? There has been considerable work to automate classification of email. Yet, it does not give a good sense of how well we are able to model user's classification of email. This paper reports the results of our own evaluations, including a stark observation that evaluation of this class of adaptive system needs to take account of the fact that the user can be expected to adapt to the system. This is important for the long term evaluation of such systems since we may find that this effect means that our systems may be performing better than classic evaluations might suggest.
Modelling Reputation in Agent-Based Marketplaces to Improve the Performance of Buying Agents BIBAFull-Text 273-282
  Thomas Tran; Robin Cohen
We propose a reputation oriented reinforcement learning algorithm for buying agents in electronic market environments. We take into account the fact the quality of a good offered by different selling agents may not be the same, and a selling agent may alter the quality of its goods. In our approach, buying agents learn to avoid the risk of purchasing low quality goods and to maximize their expected value of goods by dynamically maintaining sets of reputable and disreputable sellers. Modelling the reputation of sellers allows buying agents to focus on those sellers with whom a certain degree of trust has been established. We also include the ability for buying agents to explore the marketplace in order to discover new reputable sellers. In this paper, we focus on presenting the experimental results that confirm the improved satisfaction for buying agents that model reputation according to our algorithm.
Customising the Interaction with Configuration Systems BIBAFull-Text 283-287
  Liliana Ardissono; Anna Goy; Matt Holland; Giovanna Petrone; Ralph Schäfer
This paper presents an intelligent user interface for the management of personalised configuration in business-oriented domains. The proposed system fills the gap between the technical interaction style adopted by current configuration systems and the user's needs, by assisting the user during the selection of the features of the items to be configured and by customising the presentation of the solutions.
Does Adapted Information Help Patients with Cancer? BIBAFull-Text 288-291
  Diana Bental; Alison Cawsey; Janne Pearson; Ray Jones
Models of patients' information needs have great potential in improvinghealth information. Such models can adapt information to patients' medical circumstances, educational level and psychological needs. However, building these models and the information systems based on them can be difficult and costly, and it is difficult to assess the benefits of such systems for the patients. We describe a study to compare the psychological effects for cancer patients of tailored information against information that has not been tailored.
Empirical Evaluation of Adaptive User Modeling in a Medical Information Retrieval Application BIBAFull-Text 292-296
  Eugene Santos; Hien Nguyen; Qunhua Zhao; Erik Pukinskis
A comprehensive methodology for evaluating a user model presents challenges in choosing metrics and in assessing usefulness from both user and system perspectives. In this paper, we describe such a methodology and use it to assess the effectiveness of an adaptive user model embedded in a medical information retrieval. We demonstrate that the user model helps to improve the retrieval quality without degrading the system performance and identify usability problems overlooked in the user model architecture. Empirical data help us in analyzing drawbacks in our user model and develop solutions.
Multivariate Preference Models and Decision Making with the MAUT Machine BIBAFull-Text 297-302
  Christian Schmitt; Dietmar Dengler; Mathias Bauer
With the advent of e-commerce, systems supporting the user in finding just the right product in an electronic catalog have gained increasing attention. While collaborative recommender systems (RS) derive their suggestions from other users' opinions, structure-based systems assess a product according to how well its properties satisfy a user's preferences. This paper presents the MAUT Machine, a system implementing the basic machinery to be used by a structure-based RS to elicit and maintain complex user preference models and evaluate the entries of an electronic catalog according to their appropriateness for a given user or group of users.

Student Modeling Methods

Predicting Student Help-Request Behavior in an Intelligent Tutor for Reading BIBAFull-Text 303-312
  Joseph E. Beck; Peng Jia; June Sison; Jack Mostow
This paper describes our efforts at constructing a fine-grained student model in Project LISTEN's intelligent tutor for reading. Reading is different from most domains that have been studied in the intelligent tutoring community, and presents unique challenges. Constructing a model of the user from voice input and mouse clicks is difficult, as is constructing a model when there is not a well-defined domain model. We use a database describing student interactions with our tutor to train a classifier that predicts whether students will click on a particular word for help with 83.2% accuracy. We have augmented the classifier with features describing properties of the word's individual graphemes, and discuss how such knowledge can be used to assess student skills that cannot be directly measured.
A Comparative Analysis of Cognitive Tutoring and Constraint-Based Modeling BIBAFull-Text 313-322
  Antonija Mitrovic; Kenneth R. Koedinger; Brent Martin
Numerous approaches to student modeling have been proposed since the inception of the field more than three decades ago. What the field is lacking completely is comparative analyses of different student modeling approaches. In this paper we compare Cognitive Tutoring to Constraint-Based Modeling (CBM). We present our experiences in implementing a database design tutor using both methodologies and highlight their strengths and weaknesses. We compare their characteristics and argue the differences are often more apparent than real: for specific domains one approach may be favoured over the other, making them viable complementary methods for supporting learning.
Assessing Student Proficiency in a Reading Tutor That Listens BIBAFull-Text 323-327
  Joseph E. Beck; Peng Jia; Jack Mostow
This paper reports results on using data mining to extract useful variables from a database that contains interactions between the student and Project LISTEN's Reading Tutor. Our approach is to find variables we believe to be useful in the information logged by the tutor, and then to derive models that relate those variables to student's scores on external, paper-based tests of reading proficiency. Once the relationship between the recorded variables and the paper tests is discovered, it is possible to use information recorded by the tutor to assess the student's current level of proficiency. The major results of this work were the discovery of useful features available to the Reading Tutor that describe students, and a strong predictive model of external tests that correlates with actual test scores at 0.88.
Adaptive Bayes for a Student Modeling Prediction Task Based on Learning Styles BIBAFull-Text 328-332
  Gladys Castillo; João Gama; Ana M. Breda
We present Adaptive Bayes, an adaptive incremental version of Naïve Bayes, to model a prediction task based on learning styles in the context of an Adaptive Hypermedia Educational System. Since the student's preferences can change over time, this task is related to a problem known as concept drift in the machine learning community. For this class of problems an adaptive predictive model, able to adapt quickly to the user's changes, is desirable. The results from conducted experiments show that Adaptive Bayes seems to be a fine and simple choice for this kind of prediction task in user modeling.
User Modeling and Problem-Space Representation in the Tutor Runtime Engine BIBAFull-Text 333-336
  Steven Ritter; Stephen Blessing; Leslie Wheeler
Our efforts to commercialize Cognitive Tutors have led us to a runtime representation that is significantly different from the production system representation used in the Tutor Development Kit. This paper describes our new representation, which we call the Tutor Runtime Environment (TRE).
A Neuro-fuzzy Approach in Student Modeling BIBAFull-Text 337-341
  Regina Stathacopoulou; Maria Grigoriadou; George D. Magoulas; Denis Mitropoulos
In this paper, a neural network-based fuzzy modeling approach to assess student knowledge is presented. Fuzzy logic is used to handle the subjective judgments of human tutors with respect to student observable behavior and their characterizations of student knowledge. Student knowledge is decomposed into pieces and assessed by combining fuzzy evidences, each one contributing to some degree to the final assessment. The neuro-fuzzy synergism helps to represent teacher experience in an interpretable way, and allows capturing teacher subjectivity. The proposed approach was used to assess knowledge and misconceptions of simulated students interacting with the exploratory learning environment "Vectors in Physics and Mathematics", which is used by high school pupils to learn about vectors. In our experiments, this approach provided significant improvement in student diagnosis compared with previous attempts.

Learning Environments: Natural Language and Pedagogy

Student Modeling for an Intelligent Agent in a Collaborative Learning Environment BIBAFull-Text 342-351
  F. Linton; B. Goodman; R. Gaimari; J. Zarrella; H. Ross
We present an application of student modeling in support of individual learning and group collaboration. The design is based on an empirical analysis of collaborative dialogs collected in earlier work. During use, student modeling data will be acquired not only from learners' graphical user interface actions but also from their text chat by using keywords and speech acts. An agent in the collaborative learning environment will promote deliberative discussion and individual learning based on these group and student models.
A Teaching Model Exploiting Cognitive Conflict Driven by a Bayesian Network BIBAFull-Text 352-362
  K. Stacey; E. Sonenberg; A. Nicholson; T. Boneh; V. Steinle
This paper describes the design and construction of a teaching model in an adaptive tutoring system designed to supplement normal instruction and aimed at changing students' conceptions of decimal numbers. The teaching model exploits cognitive conflict, incorporating a model of student misconceptions and task performance, represented by a Bayesian network. Preliminary evaluation of the implemented system shows that the misconception diagnosis and performance prediction performed by the BN reasoning engine supports the item sequencing and help presentation strategies required for teaching based on cognitive conflict. Field trials indicate the system provokes good long term learning in students who would otherwise be likely to retain misconceptions.
Towards Intelligent Agents for Collaborative Learning: Recognizing the Roles of Dialogue Participants BIBAFull-Text 363-367
  Bradley Goodman; Janet Hitzeman; Frank Linton; Helen Ross
Our goal is to build and evaluate a web-based, collaborative distance-learning system that will allow groups of students to interact with each other remotely and with an intelligent agent that will aid them in their learning. The agent will follow the discussion and interact when it detects learning trouble of some sort, such as confusion about the problem they are working on or a participant who is dominating the discussion or not interacting with the other participants. In order to recognize problems in the dialogue, we are first examining the role that a participant is playing as the dialogue progresses. In this paper we discuss group interaction during collaborative learning, our representation of participant roles, and the statistical model we are using to determine the role being played by a participant at any point in the dialogue.
Modeling Student Performance to Enhance the Pedagogy of AutoTutor BIBAFull-Text 368-372
  Tanner Jackson; Eric Mathews; King-Ip Lin; Andrew Olney; Art Graesser
The Tutoring Research Group from the University of Memphis has developed a pedagogically effective Intelligent Tutoring System (ITS), called AutoTutor, that implements conversational dialog as a tutoring strategy for conceptual physics. Latent Semantic Analysis (LSA) is used to evaluate the quality of student contributions and determine what dialog moves AutoTutor gives. By modeling the students' knowledge in this fashion, AutoTutor successfully adapted its pedagogy to match the ideal strategy for students' ability.
Modeling Hinting Strategies for Geometry Theorem Proving BIBAFull-Text 373-377
  Noboru Matsuda; Kurt VanLehn
This study characterizes hinting strategies used by a human tutor to help students learn geometry theorem proving. Current tutoring systems for theorem proving provide hints that encourage (or force) the student to follow a fixed forward and/or backward chaining strategy. In order to find out if human tutors observed a similar constraint, a study was conducted with students proving geometry theorems individually with a human tutor. When working successfully (without hints), students did not consistently follow the forward and/or backward chaining strategy. Moreover, the human tutor hinted steps that were seldom ones that would be picked by such tutoring systems. Lastly, we discovered a simple categorization of hints that covered 97% of the hints given by the human tutor.

Mobile and Ubiquitous Computing

User Modelling in the Car BIBAFull-Text 378-382
  Niels Ole Bernsen
The paper presents work on user modelling of car drivers. The paper presents an implemented solution to user modelling in the car, which includes an aspect of location-based user modelling.
User Modelling and Mobile Learning BIBAFull-Text 383-387
  Susan Bull
This paper describes a study investigating the potential for two user modelling systems: a location-aware user modelling system providing easy access to applications, files and course materials commonly used by an individual student in different locations; and a mobile open learner model for consultation by a student away from the intelligent tutoring system in which the learner model was generated.
D-ME: Personal Interaction in Smart Environments BIBAFull-Text 388-392
  Berardina De Carolis; Sebastiano Pizzutilo; Ignazio Palmisano
Ubiquitous access to information services in active environments depends on the user and on the situation in which interaction occurs. We propose a multiagent architecture in which users and environments are represented by agents that negotiate tasks execution and generate results according to user in context features.
A User Modeling Markup Language (UserML) for Ubiquitous Computing BIBAFull-Text 393-397
  Dominik Heckmann; Antonio Krueger
Ubiquitous computing offers new chances and challenges to the field of user modeling. With the markup language UserML, we try to contribute a platform for the communication about partial user models in a ubiquitous computing environment, where all different kinds of systems work together to satisfy the user's needs. We also present an implementation architecture of a general user model editor which is based on UserML. The keywords are ubiquitous computing, distributed user modeling and markup languages.
Purpose-Based User Modelling in a Multi-agent Portfolio Management System BIBAFull-Text 398-402
  Xiaolin Niu; Gordon McCalla; Julita Vassileva
This poster outlines a new approach for decentralized user modelling using a taxonomy of purposes that define a variety of context-dependent user modelling processes rather than creating and maintaining a single centralized user modelling server. This approach can be useful in distributed environments where autonomous agents develop user models independently and do not necessarily adhere to a common representation scheme.
User Modeling in Adaptive Audio-Augmented Museum Environments BIBAFull-Text 403-407
  Andreas Zimmermann; Andreas Lorenz; Marcus Specht
The paper illustrates approaches for making audio-augmented museum environments adaptive. Based on well-known user modeling techniques we present a combination of suitable components that adapt audio information to the interests, preferences and motion of a museum's visitor. The underlying environment, i.e. the carrier and transmitter of information, is provided by the LISTEN system, which enables the augmentation of everyday environments with audio information.

Doctoral Consortium Papers

MAPS: Dynamic Scaffolding for Independence for Persons with Cognitive Impairments BIBAFull-Text 408-410
  Stefan Carmien
Individuals with cognitive disabilities are often unable to live independently due to their inability to perform daily tasks. Computationally enhanced dynamic prompting systems can mitigate this inability. Poor user interfaces, however, drive high levels of assistive technology abandonment by this population. To address this issue, MAPS (Memory Aiding Prompting System) provides an effective prompting system with an intuitive interface for configuration. User modeling techniques facilitate simple and effective prompting scripts for individual user needs.
Adaptations of Multimodal Content in Dialog Systems Targeting Heterogeneous Devices BIBAFull-Text 411-413
  Songsak Channarukul
Dialog systems that adapt to different user needs and preferences appropriately have been shown to achieve higher levels of user satisfaction [1]. However, it is also important that dialog systems be able to adapt to the user's computing environment, because people can access computer systems using different devices. Existing research has focused on either user-centered adaptations or device-centered adaptations. To my knowledge, no work has been done on integrating and coordinating both types of adaptation interdependently. In this thesis, I aim to investigate how multimodal dialog systems can adapt their content and style of interaction to individual users and their current device. The primary contribution of this thesis will be a framework that extends and combines both types of multimodal content adaptations that should occur in dialog systems.
Learning Knowledge Rich User Models from the Semantic Web BIBAFull-Text 414-416
  Gunnar Astrand Grimnes
The SemanticWeb [2] is a vision in which today'sWeb will be extended with machine readable content, and where every resource will be marked-up using machine readable metadata. The intention is that documents on the SemanticWeb will convey real meaning by using structured data-formats and by referring to common ontologies.
Modeling User Navigation BIBAFull-Text 417-419
  Eelco Herder
For providing users with navigation aids that best serve their needs, user models for adaptive hypermedia should include user navigation patterns. This paper describes elements needed and how these elements can be gathered.
A Longitudinal, Naturalistic Study of Information Search & Use Behavior as Implicit Feedback for User Model Construction & Maintenance BIBAFull-Text 420-422
  Diane Kelly
A longitudinal, naturalistic study of the online information search and use behavior of seven users is being conducted during a four-month period to understand how behavior can be used as implicit sources of evidence for user model construction and maintenance. Users are provided with laptops and printers, and their activities are monitored with logging software, paper instruments and weekly interviews. The goal of the study is to develop methods for using online search and use behaviors to predict document usefulness in order to unobtrusively build and maintain a model of the user's interests.
Facilitating the Comprehension of Online Learning Courses with Adaptivity BIBAFull-Text 423-425
  Stefan Lippitsch
Knowledge acquisition with texts is assumed to be a process of building a mental model of the specific subject. For readers with more prior knowledge, the building of an accurate mental model is easier because they do not have to establish a new structure. Readers with less or no prior knowledge might build an inadequate mental model of a subject. In a hypertext learning environment this could be prevented by several adaptive features that support the user with additional information. We plan to examine the effectiveness and efficiency of such adaptive features within an online course by assessing the user's acquired domain knowledge, the user's satisfaction, and achievement of the user's objectives.
Scrutable User Models in Decentralised Adaptive Systems BIBAFull-Text 426-428
  Andrew Lum
This research focuses on users being able to inspect and manage large client side user models in a decentralised adaptive system. In particular, users should be able to scrutinise and modify which parts of their model can contribute to a partial user model that shall be made public. An ontology will be used to structure the user model, allowing users to see what can be inferred about them from this partial user model and also to serve as a basis for visualisation. This work will be evaluated in an online learning context and utilise current web standards.
A Pseudo-Supervised Approach to Improve a Recommender Based on Collaborative Filtering BIBAFull-Text 429-431
  José D. Martín-Guerrero
This PhD Thesis develops an optimal recommender. First of all, users accessing to a Web site are clustered. If a user belongs to a cluster, the system offers services which are usually accessed by users from the same cluster in a collaborative filtering scheme. A novel approach based on a users simulator and a dynamic recommendation system is proposed. The simulator is used to create the situations that one can find in a Web site. Introduction of dynamics in the recommender allows to change the clusters and in turn, the decisions which are taken. Since the system is based both on supervised and unsupervised learning whose borders are not too clear in our approach, we talk about a pseudo-supervised learning.
Visualizing a User Model for Educational Adaptive Information Retrieval BIBAFull-Text 432-434
  Swantje Willms
We will visualize a user model to achieve adaptive information retrieval in a learning environment. User profile components such as user interests and user knowledge are visualized in WebVIBE with the purpose of helping the student in identifying relevant documents for study. We will be able to find out whether this visualization will change the students' access patterns to the study material and whether they will be able to find relevant documents faster.