UserModeling and User-AdaptedInteraction 6: 69-76, 1996. 9 1996KluwerAcademicPublishers. Printedin the Netherlands.

69

ABIS-95 - GI Workshop on Adaptivity and User Modeling in Interactive Software Systems Siemens AG, Corporate Research & Development, Dept. Man-Machine Cooperation, Munich, Germany, 11-13 October 1995

UWE MALINOWSKI Siemens ZFE T SN 5, D 81730 Munich, Germany e-mail: [email protected]

In October 1995, the Third German Workshop on Adaptivity and User Modeling in Interactive Software Systems (ABIS-95) took place at Siemens Corporate Research in Munich. Following the two successful workshops in Berlin 1993 (conference report in UMUAI Vol. 3 No. 4, pp. 359-367) and Sankt Augustin near Bonn 1994 (conference report in UMUAI Vol. 4 No. 2, pp. 131-138), in this year's workshop the focus was explicitly set on demonstration of adaptivity and user modeling techniques in existing systems; short presentation of innovative ideas; discussion of topics that are most important barriers on the way to massive usage of adaptivity and user modeling in commercial applications. During the workshop, 6 long presentations including system presentations, and 9 short presentations of research ideas or current projects were given. Their abstracts are given following this introduction. From the papers presented, proceedings were produced. Copies are available from Uwe Malinowski (contact see above; note that the papers are written in German). Additionally, two discussion groups worked on the topics "Adaptivity in Commercial Applications" and "Group Modeling". The main ideas from the discussions are presented following the presentation abstracts. From the presentations and discussions it became obvious that industry as well as universities are currently trying to apply research results to real world applications. The questions addressed vary from providing shell systems on different platforms to reducing complexity in inference or decision algorithms. It was generally agreed that it is necessary to reduce the cost of building adaptive systems; it is still too expensive to build an adaptive interactive system, especially if the effect of adaptivity cannot be proven in advance. The next ABIS workshop will take place in Dortmund from 9-11 October 1996. It will be organized by Hans-Gianter Lindner from VEW AG ([email protected]). Information will be accessible via the WWW homepage of the GI work-

-

-

70

UWEMALINOWSKI

ing group "Adaptivity and User Modeling in Interactive Software Systems" (http://zeus.gmd.de/GI/fg2.3.3.html).

1. Abstracts of Long Presentations HANS-GONTER LINDNER

(VEW, [email protected] ) Adaptive Literature Management using "author in a box" "Author in a box" is a Lotus Notes data base that ensures constant quality of literature management for teams at VEW. Typical users in the teams are secretaries, specialists, and managers. Adaptive facilities are given by varying document layout, consulting pop-ups with changing content, context oriented correction, hints for work flow control and adaptable group help for each team. The adaptation is based on a simple neural model where each field of the database is represented as a neuron. User actions activate neurons in each document and the activation can propagate through the network. If a neuron's activity is higher than its threshold, messages to the user will be given. A user survey shows the acceptance and the usability of the adaptive application. MIGUEL ENCARNACAO

(University of Tiibingen, [email protected]) Adaptive Help for Interactive Graphics Systems: An Application-Independent Approach Adaptive help components are an essential extension for complex systems that aim at providing usability for a broad range of users with different levels of expertise. The required adaptivity presupposes the availability of sophisticated user and task modeling techniques whose adequacy may differ among different applications and needs to be carefully evaluated. The presented framework supports the development of appropriate adaptive help systems for new and existing applications. It provides a simple user model based on stereotypes and a probabilistic evaluation of the dialog history, but is designed to allow the direct comparison of results of different user and task modeling approaches in graphical user interfaces. The application-independent implementation supports different help strategies and multimodal help presentations, and can be easily adapted to different application systems. A prototype implementation was presented providing an adaptive hypertext help component for sample medical and CAD experimentation environments. MARTIN KOHNEN, REINHARD OPPERMANN, AND CHRISTOPH G. THOMAS

(GMD ; {R einhard.Oppermann, Christoph.Thornas } @gmd.de ) Usage Modeling exemplified on Search Engines in the World-Wide Web The World-Wide Web (WWW) provides a massive amount of information. Users

ABIS-95 - GI WORKSHOP

71

get lost in space and overloaded with information. There is more information than an information consumer currently seeks and needs, and finding the relevant information can be frustrating and time-consuming. Tools like search engines help users search the WWW and to locate desired information. But this process is static: users have to filter search results according to their current preferences, interests, or tasks on their own. In this paper, we discuss our concept of supporting the search and filter process by introducing search agents combined with implicit and explicit user modeling techniques, as has been developed in the system BASAR. The search agents mediate between users and existing search engines. They dynamically filter search results according to user's specific preferences, interests, and tasks modeled in the usage profile. HANS-G~INTER LINDNER

(VEW, [email protected]) A Prototype for an Adaptive Support for Insurance Agents The adaptive insurance sales system supports the interaction between agent and customer. Adaptation with neural techniques is realized for the navigation, for offering additional insurance, and in an active help system. The navigation adapts to the agent's preferred style of selling, but can also be trained explicitly. Additional offers are made by the system depending on the context and the agent's estimate of how the customer takes risks. The adaptive help system supports the user with warnings, hints, and information in a hypertext system. The adaptation can be switched off or can be controlled intuitively by a critique component that influences the learning rate of the adaptive navigation, the thresholds for additional offering, and the help system. Conventional expert rules are used to adapt questions to the customer's context. A user survey proved the acceptance of and the demand for adaptive user support. Parts of the prototype were used by a German insurance company for enhancing their applications.

DO-WAN KIM

(University of Regensburg, [email protected] ) WING-MIT: A Tutorial Help System based on an Intelligent Multimedia User Interface The tutorial help-system WlNG-MIT (Multimedial and Intelligent Tutorial HelpSystem for WING-M2) is a hybrid between a help-system and a tutoring system. Therefore it has two goals: firstly, it provides necessary help without interfering with the user's goals, and secondly, it facilitates a high learning effect. These objectives are achieved in WlNG-MIT by integrating intelligent components like a plan-recognition component and a user modeling layer into the system, thus modeling a human tutor who also helps efficiently by combining a help and a tutor mode. WING-MIT is designed in such a way that the user will not only find the structure of the system in use, but also the problematic situation

72

trWEMALINOWSKI

which lead him to using WING-MIT. It is explained to the user what he had done (dialog history), why the error occurred (context-sensitive diagnosis) and how he can solve the problem (context-sensitive help). Help information is presented in a multimedial, adaptive, and context-sensitive way using intelligent components. Characteristics of WING-MIT are the teaching-learning-process, possibilities for exploration, heuristic choice of training sessions, and the symbiotic overall design of the complex system. ALASSANE NDIAYE AND ANTHONY JAMESON

(University of Saarbriicken, { ndiaye, jameson }@cs.uni-sb.de ) Global Anticipation Feedback in a Transmutable Dialog System Human dialog participants regularly predict the responses of their dialog partners by hypothetically assuming the partner's role. This strategy of using "global anticipation feedback" has seldom been approximated in dialog systems. In the system PRACMA (see, e.g., the Proceedings of IJCAI'95, pp. 1886-1893), the technical prerequisites for this strategy have been fulfilled, and the system is being used as a testbed to explore the potential and limitations of the strategy. This paper discusses several implemented variants of the strategy which differ in terms of their computational cost and the extent to which they deal with uncertainty and with differences between the system and the user. (An expanded version of this paper is available in the Proceedings of UM'96 and from http://zaphod.cs.uni-sb.de/KI/) 2. Abstracts of Short Presentations

MATJAZ DEBEVC, RAJKO SVECKO, DALI DONLAGIC

(University Maribor, [email protected] ) A Simplified Decision Making Component for Adaptive User Interfaces The majority of adaptive mechanisms for adaptive user interfaces are based on three different models: a user model, a task model, and a system model. The more complex the adaptive user interface is, the more difficult it is to design and create these models. This paper presents a decision-making component that simplifies the design of adaptive mechanisms without large data bases. The decision-making component is based on monitoring the users' past actions. An example of the effect of that mechanism has been implemented for Microsoft Word for Windows in the form of a special adaptive icon bar, called the Adaptive Bar. The paper describes the criteria that the system uses to select an appropriate icon to offer in the bar. In this example, the adaptive mechanism works very fast in the background so that the user is not interrupted when working. In conclusion, the adaptive bar improved user performance when experimentally tested along with a non-adaptive version.

ABIS-95 - GI WORKSHOP

73

KNUT PITSCHKE

(University of Oldenburg, [email protected]) Classification of Approaches to User Modeling and Implications for Explanation Generation Existing approaches to user modeling are classified according to the domain knowledge represented in the system, and to the kind of knowledge about the user available in the system. From this, implications are derived for the kinds of explanations that can be generated utilizing the different kinds of user models. Finally, it is shown how explanations are generated in Petri-Help, an Intelligent Tutoring System for the domain of petri-nets, which uses a probabilistic model of students' behavior to predict observable actions, and to generate help information. GERHARD PETER (FAWUlm, [email protected]) A Concept for Intelligent User Support in the Project IPQM This paper describes how a heterogeneous group of users working in the area of preventive quality assurance can be supported. The analysis of the group, its members and the application motivates the introduction of the following models: group model, user model, task model, and role model. Each model is briefly described and the relationships between these models are explained. The paper concludes with a discussion of the approach and emphasizes future research plans. STEPHAN ROMAHN

(RWTHAachen, [email protected]) Automated Classification of Pilot Errors in Flight Management Operations This paper introduces SmartTranscript, a tool for automated analysis of pilot action protocols recorded while operating the Flight Management System (FMS) of modem commercial aircraft. Since errors are valuable indicators of pilots' mental workload and therefore of the user interface quality, SmartTranscript checks pilot action protocols for occurred user errors. The conceptual model of operator's tasks, proposed by Rouse & Rouse (1983), together with its human error categories, was found suitable to classify pilot errors detected in experiments. In addition to traditional approaches, SmartTranscript is a means to evaluate pilot-FMS interfaces objectively and enables comparison of interface alternatives. SAHIN ALBAYRAK, ULRIKE HAMMER, OLIVER KUNDT, EBERHARD STEINBAUER

(Technical University of Berlin, [email protected]) The Service Adaptivity for the Arrangement of Information and Communication Technologies in Public Administration The need for a service called Adaptivity and its possible application in public administration is pointed out in this paper. The integration of complex work flows in distributed environments demands for groupware and workflow management systems. Adaptivity has to support these in the same way as applications of telecom-

74

uwE MALINOWSKI

munication and usual office applications. Therefore, domain independence of the service is required. A practicable construction with two agents is demonstrated. One of these is a desktop agent which represents the interface to other services and applications. Therefore the desktop agent depends on the domain. It performs the static processing, e.g. configuration. The diagnosis agent is independent of the domain and responsible for the dynamic processing. Through the distribution of the functions to two agents the request of domain independence obtains particular consideration. RALPH SCH~ZER

(University of Saarbriicken, [email protected] ) Information Value as a Criterion for the Selection of Queries Concerning User Characteristics A system that presents evaluation-oriented information to the user should be able to assess the user's evaluation criteria (cf. the Proceedings of IJCAI'95, pp. 18861893). There are various ways in which the system can elicit from the user the information on which to base such an assessment. For example, the system can ask the user to supply information about personal characteristics such as his or her profession; or it can evoke explicit evaluative reactions. But such methods must be applied selectively. It is therefore important for the system to be able to predict in advance the value of the information that will be provided. This paper illustrates how this can be done if the system uses Bayesian networks as a framework for modeling the user's evaluations. MARCUS SPECHT, GERHARD WEBER, PETER BRUSILOVSKY

(University Trier, [email protected] ) Episodic User Modeling for Interface Adaptation In most classical intelligent learning environments (ILE), merely the tutoring component is adaptive to the users skills and the knowledge in the domain being taught. In the ELM-PE project we try to adapt all components of an ILE to the user. Adaptation can be based on several levels of knowledge: conceptual, procedural (goal-plan oriented) and episodic. ELM-PE uses a particularly powerful episodic learner model (ELM). The episodic model allows an adaptation on all the knowledge levels introduced above. For each of these levels we suggest different methods for adapting the interface in ELM-PE. With the ELM we can achieve an adaptation to users in different stages of learning and to individual episodes of their leaming history. JOSEF FINK

(GMD , Josef.F [email protected] ) An Open and Flexible Architecture for User Modeling Shell Systems Many of today's user modeling shell systems have an architecture that is based on a one-to-one relationship between adaptive application and user modeling system.

ABIS-95 - GI WORKSHOP

75

Therefore, they can hardly address interesting future scenarios in the field of user modeling where synergetic effects arising from the distributed use of a central user model might play an important role. We propose a generic architecture for the user modeling shell system BGP-MS, which has been designed to cope with different possible user modeling scenarios and at the same time offers a high degree of transparency that is manifested in a uniform view of the available functionality. WOLFGANG POHL, JORG HOHLE

(GMD,[email protected]) BGP-MS-jr. - a PC Port of BGP-MS and Its Applications User modeling is a technique aimed at helping users master complex software applications by having the applications adapt to their users' knowledge and needs. In order to support developers of a wide range of applications and to gain more experience about the techniques of user modeling, part of the user modeling shell system BGP-MS has been ported to the MS-Windows environment, resulting in BGP-MS-Jr. This is a restricted version of BGP-MS, but offers special mechanisms for an interesting range of applications. Experiences with two applications, an adaptive hypertext and an intelligent tutoring system, show that the restricted system is usable and can be extended according to special demands of an adaptive application.

3. Summary of Discussion Groups Adaptivity in Commercial Applications reported by UWE MALINOWSKI

(Siemens CorporateResearch,[email protected]) This group discussed questions concerning the development of commercial interactive applications with adaptive components. The following criteria describe possible advantages that can be achieved with an adaptive component (compared to the respective non-adaptive component): systems can be used more efficiently; less errors are made while working with the system; higher quality of work results; less teaching and support necessary; higher acceptance by the user. At the same time, the following aspects need to stay at least at the same level: flexibility; maintainability; transparency and control. When transferring results from the research lab to real world applications, the following aspects require special consideration: predictability and evaluation of the effect of adaptive components; cost of implementation; misuse bf user models for performance control; anonymization of user models; acceptance by managers and union; profits from adaptation; and adaptivity vs. consistency. Considering the most important existing constraints (cost of implementation; good maintainability, transparency, and control; excellent user acceptance) in the short run the following features seem to be the most promising candidates for adaptation in commercial applications: toolbox adaptation; automation of simple tasks; providing task-sensitive information at the right time.

76

OWEMALINOWSKI

Group Modeling reported by WOLFGANG POHL ( GMD , Wolfgang.P [email protected] ) Group modeling might be an important method for intelligent computer systems that are supposed to support and even cooperate with groups of users. In this forum, systems for which the use of a group model could be imagined were discussed. Among these are cooperation assistants for heterogeneous work groups and group teaching and learning systems. Based on the latter scenario, the potentials of group modeling were examined. In a subsequent general discussion several statements were agreed upon: Group models will mainly support the communication between group members. This support is probably most effective if there is some degree of heterogeneity within the group (like in a class with the differing individual abilities of the students). The following information should be stored in a group model: knowledge common to all group members, which can be derived from individual user models and may be used to ensure the use of a common vocabulary in group communication; relationships between and roles of group members or subgroups; group goals (in case a group was formed to achieve one or more common goals) whose dynamic modeling might support group decision processes; and a common base of rules of group behavior that should be considered by group members for the sake of a fruitful cooperation. Topics related to intelligent cooperation with and within groups have already been raised in other research areas like organizational science, social science, psychology, and CSCW. It must be investigated whether the obtained results are useful and inspiring for research on the use of group models in adaptive computer systems.