Edited Adaptive Hypermedia: Combining Human and Machine Intelligence to Achieve Filtered Information Kristina Höök SICS
[email protected]
Åsa Rudström Stockholm University
[email protected]
Annika Waern SICS
[email protected]
(To be presented at the workshop "Flexible Hypertext" to be held in Southampton during the Hypertext conference in April 6 11th, 1997.)
Abstract We discuss a novel approach to filtering of hypermedia information based on an information broker and user environment coupled together. The advantage of the proposed approach, edited adaptive hypermedia , is that it combines human expertise with machine intelligence in order to achieve high quality of the filtered information provided to the end users.
1. Background and motivation Lately, adaptive hypermedia has attracted a lot of attention, as a means to tackle the problems users encounter with information overflow and navigation through large information spaces and ordinary hypermedia (Brusilovsky 1996, Höök et al. 1996). Adaptive hypermedia takes into account the fact that users vary in knowledge, cognitive skills and reasons for searching for information. By keeping a model of some aspects of user characteristics the system can adapt to and aid the user to navigate and filter information. A major difficulty in producing adaptive hypermedia systems, or indeed user-adaptive systems in general, lies in structuring the information in such a way that it will be possible to do adaptations. The representation must include characterizations of users that allow for useful adaptations, and the interface must be structured to allow the underlying system to infer the required characteristics from user actions at the hypermedia interface. This problem is most apparent in domains where the information is rapidly changing or highly unstructured. How could we, for example, analyse and represent the widespread needs of users of the WWW in such a way that it would be possible to filter information or adapt navigation to an individual user? And even if we could, how could we infer those needs from just observing the user's navigation through the WWW? Promising approaches to this problem are those where the user community itself provides the needed structure through their preferences and actions. The structure can be provided directly by the user (filling in keywords, setting rules in e.g. email filters etc). However, according to e.g. Schneiderman (1987), users will not perform actions that will not render immediate gain to themselves. In most cases, indirect, automated approaches are to be preferred, since they do not require any extra actions from the end user. The system just observes the choices made by users and tries to infer their underlying goals or learn about their preferences. Pattie Maes, (1995) argues that one reason that software agents should be used is that users sometimes do not want to bother with detailed tasks, and instead will want to delegate them to a separate agency. For example, we are usually quite willing to delegate to a car mechanic to fix our car, even if this means that we do not quite understand what has been fixed and how it was done. Most approaches to inferring user preferences are based on the actions of a single user. A problem with such approaches is that it will be hard, if not impossible, for the system deal with new information. The Firefly system, developed at MIT Media Lab, takes care of this problem. The preferences of an individual user are compared to those of the full user society, and the user is grouped together with others expressing
roughly the same preferences. This way, the system is able to suggest new information based on the fact that other users with a similar preference pattern have liked this information. However, it may not always be the case that the preferences of a whole group of people will be able to satisfy a particular user's needs. In fact, that user might be much more interested in what a single expert would regard as important information, rather than in the recommendation of a large group of peers. In the general case, users may want to judge the relevance of a piece of information based both on quality (the expertise of the recommender(s)) and on quantity (the number of people recommending it). There is also the question of trust, as discussed by Maes (1994). Experiments (e.g. Bonsall and Joint (1991)) show that users have difficulties in placing the right level of trust in automatised systems. Initially, the trust is often too high, but lowers dramatically once the system makes a single error. It is much easier to know what to expect if the service is provided by another human being rather than an automated agent. Our solution to the above problems is to put the human editor back in place. We want to combine the skills of professionals with machine intelligence in order to filter information and get feedback on user preferences. In particular, we want to focus on the structuring and authoring of adaptive hypermedia, a problem not much discussed in literature (Höök 1996). The next section outlines our vision of such an integrated environment, while the involved problems and suggestions for their solution are discussed in section three.
2. Vision Our vision is to put the human editor back in place. An editor, or information broker, is a person, such as a journalist, publisher, scientist or librarian, or just a dedicated individual, who collects and structures information for the benefit of other information users. Usually, the information broker has specialist knowledge in a subject, and knows more than others about how to find and evaluate information on this subject. The information broker will have a more or less clear picture of what his/her customers (newspaper readers, book readers, other scientists, etc.) want and will adapt to their needs. Information brokers collect information from various sources, evaluate its relative importance and then choose whether to include the information as it is, disregard it, summarize it, or perhaps rewrite or illustrate it differently than in the original source. Examples of the information broker role are professional editors and journalists that direct their services towards the open public, and managers of internal or external information within an enterprise. Information brokers already exist on the web. In many cases, these services are maintained by dedicated individuals, rather than proffessional editors. Some examples are Adaptive Hypertext & Hypermedia home page Jacob Nielsens personal site The Data Mine Case-Based Reasoning in the Web. We propose a service infrastructure that builds upon and extends the information broker scenario as it can be seen on the web today. It does so through providing support for the development of adaptive hypermedia: edited adaptive hypermedia .The service involves two types of actors, information brokers and information users, with their respective tasks of collecting, adapting, and reading the information. We suggest a solution where individual user interests and preferences are stored in user profiles, available both to the information broker and to the information user. The user profile will be split into one public and one private part as suggested by (Cook and Kay, 1994). Information collection and processing is based on clusters of such profiles. The outgoing information is annotated as to allow for individual adaptations for the information user. Finally, the information user's reading behavior is monitored and feedback is provided to the information broker, again through the user profile. The resulting architecture is shown in figure 1.
Figure 1. The Service Architecture. Edited adaptive yypermedia requires that good environments are provided both for information brokers and for information users. The information broker needs to be equipped with: an integrated environment for searching for information, utilising a wide variety of tools including instructable, learning agents for searching, selecting, restructuring/rewriting and annotating information for information users feedback from information users on their reading pattern, preferences, and understanding of the information provided. Whereas the information user can be experienced or inexperienced in using the service or in using computerised media at all, the information broker is always an expert user. Information brokers can learn to use a wide variety of tools, and can acquire advanced interaction methods for instructing the agents. They can also be provided with advanced visualisation tools such as Spotfire (Ahlberg and Shneiderman 1994) for reviewing the acquired information, both information retrieved through search, and the feedback information from users. Information users can of course be provided with the same type of environment as information brokers, but the advantage of the broker / user partition is that this is not neccessary. The information user environment can provide much simplified interaction models, and in particular, information users will not need to instruct search agents themselves.
3. Challenges 3.1 User modelling issues We propose to use user models in two ways in the edited adaptive hypermedia service. Firstly, user models
These two types of models will interact in complex ways. An obvious interaction is that the end user environment only can adapt using such annotations that the information broker has provided. In a closed information domain, an appropriate selection of annotations can be decided upon in advance, but this will not be true in general for information broker services. Instead, brokers must be provided with feedback on how well the selected annotations worked in practice. To allow for dynamic restructuring, some information about the users' profiles and reading patterns must be passed back from the information user environment to the information broker environment. This raises important privacy issues: what information can be passed on, and how do we ensure that users are in control of what information is distributed about them? This becomes particularly critical if information brokers, in turn, can exchange information about their user groups and their reading patterns. We propose to divide the user model into one part which is private and one that is public, see also (Cook and Kay, 1994). The dynamic restructuring of user characteristics and annotations require that information brokers are provided with useful tools for rewiewing and restructuring information about users. In part, these tools can be automatic, but we believe that information visualization tools will provide useful support in this task. 3.2 An integrated editor environment The editor environment must provide support for a number of editorial tasks: information search and retrieval, information visualisation, selecting, restructuring and annotating information with metadata keys generation of adaptive hypermedia spaces management of user feedback and user preferences information. For several of these tasks, useful commercial or freely available tools are already available (e.g. Metacrawler, Letitizia, Spotfire, Altavista, BASAR), or high quality research is being conducted, but the editor environment poses high demands on the integration of several tools so that they all are accessible within the same environment, and can exchange information with each other. We aim to provide an open and extendable software architecture, where different high-quality tools for the editorial tasks can be included. A fairly novel requirement though, is the need for highly usable tools for the generation of adaptive hypermedia: editing, structuring, and annotating it. The problem is quite difficult, if, for example, the adaptation happens through a stretchtext technique (Brusilovsky, 1996, Boyle and Encarnacio, 1994), the author must understand the adaptation mechanism in quite some detail before s/he can enter new information. In our view an authoring tool should aid the writer to enter text, pictures, etc. while minimising the requirement on his/her understanding of the system. Otherwise the cognitive load on the writer will be tremendous (Höök, 1996). The authoring process is difficult enough anyway.
4. Summary and concluding remarks We have discussed a novel approach to filtering of hypermedia information based on an information broker and user environment coupled together. The advantage of the proposed approach, edited adaptive hypermedia, is that it combines human expertise with machine intelligence in order to achieve high quality in the filtered information provided to the end users. A similar approach to information brokering is taken in the COBRA project, but they do not include feedback from the information users. This approach solves a number of problems related to automated filtering of information: Trust: the delegation of tasks to a software agent requires that the user trusts that the agent will do the right job. In the information broker service, the usage of a human editor will allow users to trust the system at the level they trust the professional editor responsible for the service. Novel information: using a software agent for search or filtering of information will face problems in the case when
novel topics turn up, that the user does not know of in advance and cannot declare an interest in. A human editor can in this situation act pro-actively, and redistribute this information based on his or her own judgement. Added value: The information broker uses his or her professional competence to add value to the information search: he or she can restructure, comment, illustrate or rewrite the information to fit the targeted user population.
5. References C. Ahlberg, B. Shneiderman (1994) Visual Information Seeking: Tight Coupling of Dynamic Query Filters with Starfield Displays , in Proceedings CHI'94: Human Factors in Computing Systems, pages 313 -317. Bonsall, P. W., and Joint, M. (1991) Evidence on Drivers' Reaction to In-Vehicle Route Guidance Advice , 24th ISATA International Symposium on Automotive Technology and Automation, Florence, Italy, 1991. Boyle, Craig, and Encarnacio, Antonio O. (1994) Metadoc: An Adaptive Hypertext Reading System , User Models and User-Adaptive Interaction, UMUAI 4, pp. 1 - 19. Brusilovsky, P. (1996) Methods and Techniques of Adaptive Hypermedia , Journal of User Modeling and User-Adaptive Interaction, UMUAI 6. Cook, R. and Kay, J. (1994) The Justified User Model: A Viewable, Explained User Model, Proc. of the Fourth International Conference on User Modeling, Hyannis, Mass., The Mitre Corp. Höök, K. (1996) A Glass Box Approach to Adaptive Hypermedia , Ph.D. Thesis, SICS Dissertation Series 23, ISBN: 91-7153-510-1, Stockholm, Sweden. Höök, K., Karlgren, J., Waern, A., Dahlbäck, N., Jansson, C-G., Karlgren, K. and Lemaire, B. (1996) A Glass Box Approach to Adaptive Hypermedia , Journal of User Modeling and User-Adaptive Interaction, UMUAI 6. Maes, P. (1994) Agents that Reduce Work and Information Overload , Communications of the ACM, Vol. 37, No.7,pp. 31-40, 146, ACM Press, July 1994. Maes, P. (1995) Intelligent Software: Programs that can act independently will ease the burdens that computers put on people, Scientific American, Vol. 273, No. 3, pp. 84-86, Scientific American, Inc., September 1995. Shneiderman, B. (1987) Designing the User Interface: Strategies for Effective Human Computer Interaction , Addison-Wesley.