A glass Box Approach to Adaptive Hypermedia
Doctoral thesis Stockholm University ISBN: 91-7153-510-1 Copyright © 1996 by Kristina Höök. Printed by Akademitryck AB, Edsbruk, 1996
Information overflow and navigation in large hyperspaces is becoming a familiar problem to a growing user population. Adaptive hypermedia can be used to improve users’ information retrieval behaviour and reduce the information overflow problem. In our adaptive hypermedia prototype, POP1, designed for a large Intranet database, we use stretchtext and hotwords in order to hide some information from the users’ immediate view. We infer users’ information seeking tasks from their interactions with the system, and base our decision on what to hide on what is required by their task. This enables us to reduce the amount of information and only display the most relevant information. The design of the system is based on empirical studies of the users and their tasks. Our approach to adaptivity is primarily directed at reducing the problem of information overflow. At the same time, we are concerned with how to make the adaptive behaviour understandable to the user, and give the user control over the adaptive mechanisms. There should be a balance between user- and system-controlled adaptivity. Our design basis is the ”black box in a glass box” metaphor: we hide the complex behaviour of the adaptive system in the black box and show a fairly simple model to the user in the glass box. The resulting system has been evaluated in a controlled, comparative, study, and the results show that the adaptive system was preferred over the non-adaptive variant by the subjects. Furthermore, it required fewer actions within the page, and the choice of information made by the system influenced subjects’ solutions. Keywords: Adaptive hypermedia, user modelling, user-centred design, empirical evaluation, task adaptation, computer-aided adaptation, self-adaptive system PrefaceIn January 1993 I was sitting in a rented car travelling through the Everglades national park in Florida. Together with me in the car were Nils Dahlbäck from Linköping, and two of my colleges from SICS, Annika Wærn and Jussi Karlgren. We had just attended the Internal Workshop on Intelligent User Interfaces (IWIUI) held at Disneyworld (!) in Orlando. During the trip through the Everglades, we were discussing intensively and with quite some emotional involvement, what the research problem(s) in the PUSH (Plan- and User Sensitive Help) project was and should be. The deadline for applying for money from NUTEK was close, and we wanted to approach (the former) Ellemtel Utvecklings AB2 with our ideas to see if they wanted to co-operate and contribute to the project. After seeing some of the systems presented at the IWIUI conference, we saw that the time was ripe to introduce some usability demands on the intelligent interface systems and move them from their laboratory settings, solving toy-world examples. We also saw a lack of empirical evaluation of the usefulness of adaptive systems. We decided that we wanted to build an intelligent help system for a real-world, industrial, problem. We wanted to end the project with a comparative study were we would study how much the intelligent, adaptive parts of the system were contributing with in solving the help problem. We wanted the system to have a rich interface, both graphics and text. In January 1993 we did not know that the WWW-revolution was about to take place, so when imagining the interface to the help system, we did not picture an adaptive hypermedia system with a WWW-interface, which is what we ended up with three years later. Some of the issues and ideas that came up during the car trip through the Everglades and during the following discussions back in Sweden with Prof. Carl-Gustaf Jansson and Rolf Leidhammar at the former Ellemtel Utvecklings AB, were interesting enough to cause NUTEK and Ellemtel to, together with SICS, put up the necessary funding for the project. The problems we set out to investigate lead to the research results presented in this thesis. Of course, the ideas we put into the PUSH project had their roots in the experiences and results that members of the project brought with them rather than invented in the car in the Everglades. I brought experiences mainly from the Prometheus project where I together with Annika Wærn, Carl Brown, Per Lindevall and Jussi Karlgren, designed and implemented a route guidance prototype. The route guidance system presented the routes differently depending upon the role and expertise of the driver (commuters, taxi drivers, and tourists would get different route descriptions). Carl-Gustaf Jansson brought his interest in utilising artificial intelligent in interfaces. Annika Wærn brought her plan inference ideas and experiences from the Prometheus project. Jussi Karlgren brought ideas for natural language interfaces to information retrieval tools. Nils Dahlbäck brought both his expertise in empirically-based natural language interface design and his general competence in cognitive psychology. During the years of the PUSH project, other researchers and students joined the project, each contributed ideas and helped in keeping the discussion from the car in the Everglades alive. Benoit Lemaire came with fresh ideas from his post-doc period with Johanna Moore. He convincingly showed us, both in discussions and through practical implementation that we should not try to imitate human-human communication but rather ”use computer for what computers are good at”. Catriona McDermid and Anna-Lena Ereback contributed expertise in how to perform the initial interviews with the targeted user group at Ellemtel. Klas Karlgren added ideas of how people learnt the domain we studied. Towards the end of the project, Åsa Rudström came with interesting ideas for machine learning. Malin Bladh, Fredrik Espinoza, and Marie Sjölinder who completed their Master’s theses within the PUSH project, each put in many working hours and were unusually good at making their voices heard in the discussions, thereby adding many valuable ideas. At the former Ellemtel, Måns Engstedt and the other developers of SDP contributed their expertise. STEFAN Zemke?? Now, at the end of the PUSH project, I can see that tackling a real-world problem is both rewarding but also more demanding than I expected. There were many bumps in the road and unexpected challenges, but I am proud to see that we managed to reach the two goals outlined when we started the project: to implement an adaptive help system for a real-world industrial domain, and to evaluated the adaptive parts of the system in a comparative study (even if neither the system nor the study looks like what I expected). AcknowledgementI am deeply indebted to all the members of the PUSH project mentioned above, without whom the PUSH project would not have generated the results I needed to complete this thesis. For helping me in writing this thesis, which is based on the results of the PUSH project, there are some people that I especially would like to thank. First of all, I am deeply indebted to my supervisor at the department of Computer and Systems Sciences, Carl-Gustaf Jansson. Without his never-ending enthusiasm and faith in me I would never even have started to write this thesis, even less finished it. I would also like to thank my ”extra” supervisor, Nils Dahlbäck. He has through friendly but hard criticism turned all fuzzy thoughts into clearer ideas - any fuzziness left is all my fault for not listening properly! Besides patiently reading and re-reading my thesis drafts he has also contributed many of the ideas presented. In particular, he worked together with me and Marie Sjölinder on the study of cognitive abilities and their relation to efficient usage of hypermedia. Annika Wærn has through the years not only become the person with whom I worked most, but also a dear friend. Most of the important ideas in this thesis have come from her or in discussions with her. Annika’s sharp intellect has always amazed me and I am grateful for being allowed to work with her! Before leaving us and going to New York, Jussi Karlgren was the other person at SICS with whom I could talk about ideas. Jussi has an enormous capacity for generating ideas, and he quickly understands anything I try to say. Without Annika and Jussi it would not have been possible nor particularly good fun to work so hard with the PUSH project and with my thesis writing. Thank you both! Thanks also to Annika for reading this whole thesis in detail and providing so many insightful and useful comments! I am also grateful to Åsa Rudström who read and commented my thesis, and to Torkel Franzén for reading and commenting on my horribly bad English. Also thanks to Fredrik Espinoza for the screendumps from the POP system. At one point, I desperately needed funding in order to complete my thesis writing. I am grateful to Klas Barklöf at NUTEK for providing it. I am grateful to Nomos Management AB for lending us their usability laboratory for the last study in the PUSH project. In particular I want to thank Dr. Robbin Battison and Richard Whitehand at Nomos. I would also like to thank all the anonymous subjects from Ellemtel Utvecklings AB that participated in the studies despite their heavy workload at Ellemtel during this time period. Thanks to the present and former members of the HUMLE group at SICS for providing a stimulating environment, and to Marianne Rosenqvist for being such a nice cookie general. Thanks also to my parents Evy and Gunnar Höök, parents-in-law Hillevi and Jan-Christer Janson, and friends, especially Anna-Lena Johansson, for being a support to me and my family during this period. Finally, three persons have had a rough time during my thesis writing, my husband Sverker and my children, Adam and Axel. Thank you Sverker for giving me the time, support, encouragement and love that I badly needed during this period. My lovely, beautiful children, Adam and Axel: thanks for providing all the useful distractions and for reminding me of what makes life worth living!
Upplands Väsby, August 1996 Preface and Acknowledgement i Contents v A Glass Box Approach to Adaptive Hypermedia 4 Introduction 1 Research Challenges for Adaptive Systems 2 Contributions of This Thesis 4 What This Thesis is Not About 7 Publications and Co-operation 7 Guide to The Thesis 9 A Framework for Adaptive Systems 13 What are Adaptive Systems? 14 Architecture of Adaptive Systems 15 Defining User Models 16 Classifying User Models 25 User Model Acquisition and Representation 26 Control Over the Adaptations 29 When to Adapt 30 Summary of Dimensions of Adaptive Systems 30 Adaptive Hypermedia 32 Content Adaptivity 34 Navigation Adaptivity 34 Examples of Adaptive Hypermedia Systems 35 Classifying POP 37 POP’s User Model 37 POP Is an Adaptive Hypermedia System 39 Knowledge Acquisition 41 Target Domain and Project 42 The Target Domain: SDP 42 The PUSH Project 45 Methods for Analysing Users’ Needs 46 Initial Interviews With Users of SDP 51 Method 51 Using SDP 52 Available Help Sources 52 Summary 53 Task Analysis and Corpus of Questions 53 Method 53 Results 55 Task Analysis 55 The User’s Background Knowledge 57 Users’ Roles 59 Graphic Lovers / Graphic Haters 60 Formulation of Queries 60 Users’ Questions 60 Answers 63 Questions not Gathered 64 Summary of Design Demands 64 Individual Differences and Navigation in Hypermedia 65 Method 67 Results 69 Patterns of Cognitive Abilities 69 Cognitive Ability and Task Completion Time 71 Correlation with Map-Reading Ability 72 Confidence and Efficiency 73 Discussion 75 Implications for Interface Design 75 Learning SDP Concepts and Principles 77 Method 78 Results 78 Conclusions for Design 78 Summary of the Knowledge Acquisition Phase 79 Design 81 Example Scenario 81 Non-Adaptive Scenario 81 Adaptive Scenario 85 Design Basis 86 Intelligent Tutoring Systems 87 Plan Inference (in Natural Language Systems) 90 Explanation Generation 92 Our Viewpoint: The Glass Box Model 95 Basing the Adaptivity on Users’ Tasks 100 Why Adaptive to Task? What Not to Adapt to? 100 Making Explanations Fitted to the Task 105 The Set of Information Entities 108 The Authoring Problem 112 How To Identify the User’s Task? 114 Textuality Principles 116 Where is the Glass Box? 119 What Is Unique? 120 Implementation 123 The Interface 123 An Interactive WWW Interface 123 Multimodal Input and Output 133 Summary of Interface Design 134 System Architecture 135 Knowledge Representation 136 Explanation Operators 139 Plan Recognition 139 Page Generator 140 Strengths of the Implementation 141 Bootstrapping and Evaluation 143 Evaluation of Adaptive Systems 143 Bootstrapping the Adaptivity 144 Paper and Pencil Exercise 145 Interface Evaluation and Bootstrapping of Adaptivity 146 Method and Subjects 146 Evaluation of the Interface 147 The Relevance Rules 147 Comparative Study of Adaptivity 148 Method and Subjects 148 Results 149 Navigation 150 Quality of Answers 151 User Satisfaction 152 Time Spent 154 Summary of the Comparative Study 155 Discussion 156 Concluding Remarks 157 The Problems Revisited 158 Active but Not Magical 158 Continuous Improvisation 159 Scale up 159 Imitating Human-Human Communication 160 Future Work 160 Future Work on POP 160 Figure V. A new architecture for POP.Future Work on Adaptive Hypermedia 161
Directory: ~kia -> papers ~kia -> Usability and fun An overview of relevant research in the hci community papers -> Comparing Two Approaches to Context: Realism and Constructivism ~kia -> B. Cahour (1), P. Salembier (1), Ch. Brassac (2), J. L. Bouraoui (1), B. Pachoud (3), P. Vermersch Download 0.88 Mb. Share with your friends: |