|
Åsa Rudström Computer & Systems Sciences Stockholm University/KTH Electrum 230 164 40 Kista, SWEDEN Tel. +46 8 16 1606 Fax +46 8 703 90 25 asa@dsv.su.se |
Annika Waern SICS Box 1263 164 28 Kista, SWEDEN Tel. +46 8 752 1514 Fax +46 8 751 7230 annika@sics.se |
Kristina Höök SICS Box 1263 164 28 Kista, SWEDEN Tel. +46 8 752 1517 Fax +46 8 751 7230 kia@sics.se |
Abstract: We propose a new approach to edit and adapt information from the web. The advantage of the proposed approach, edited adaptive hypermedia, is that it combines human expertise with machine intelligence. A user-adaptive WWW browser communicates with an information broker environment, in which information can be retrieved and structured to fit the target user group or groups. We exemplify the approach with an application at a Swedish company: newsletter distribution on an Intranet. The information is adapted at the reader side, based on annotations on each article from the newsletters.
Keywords: adaptivity, Intranet, information brokering, edited hypermedia.
Adaptive hypermedia has been proposed as a means for tackling problems that 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 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.
However promising, approaches to adaptive hypermedia have to address a number of issues. One major difficulty lies in structuring the information in such a way that it will be possible to do adaptations. The representation must include characterisations 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?
An alternative approach is to require that users be explicitly involved in creating the model of their preferences and behaviour. If the user has to provide this information manually, we would expect the information to mirror the user reasonably well. However, as argued by Shneiderman (1987), users are reluctant to perform actions that they will only gain from in the future. An automated solution, where the system silently observes user behaviour and learns from user actions would then seem preferable. However, observable user actions tend to be very low-level, such as mouse-clicks and menu choices, and may in particular provide weak information about what information they really seek (Waern 1997). If the system is learning from a single user's inputs, it might also take very long to learn anything useful. To remedy this problem, groups of users can contribute to each other's profiles, filling in gaps, and providing new users with an initial set of preferences. A prerequisite for this is of course that users are willing to share their profiles with other users.
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 computer systems. Initially, the trust is often too high, but lowers dramatically once the system makes a single error. Users may find it easier to place an appropriate level of trust in an information service where another human, not an automated agent, is responsible for selecting relevant information.
Finally, even if an automated agent can succeed in selecting information about topics that are relevant to a particular user, there is still a risk that the retrieved information will be of little use. The quality and relative importance of each piece of information is hard to assess, and the mere amount of information might still be overwhelming.
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). In the next section, the approach is discussed in more detail, and section three describes the application of the approach to an Intranet setting.
The EdInfo project (Edited Information Services) (Höök et al. 1997) aims at developing an environment where a human editor is supported in the task of finding, refining and distributing information to a set of readers, and where these readers are supported in their reading task. Feedback on reader interests and preferences is provided from reader to editor, so that the selection and structuring of information can be improved in the future. We refer to this as edited adaptive hypermedia.
Our approach, depicted in Figure 1, involves two types of human actors: readers and editors. 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 (but not necessarily), 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 or her customers (newspaper readers, book readers, other scientists, etc.) want, and will use this knowledge both to select and to structure the retrieved information. 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.
Individual user interests and preferences are stored in user profiles, available both to the information broker and to the reader. The individual profile will be split into one public and one private part as suggested by (Cook and Kay, 1994), to give the user control over what information is made available to the broker. Ideally, the user profile should be maintained locally on the user's machine, and only the public information should be distributed to the information broker. The information broker will base information collection and processing on group profiles, i.e. clusters of similar individual profiles. The similarity might be related to a documented interest in the same kind of topics, but also to properties such as same or similar affiliation, similar browsing behaviour, etc.
Whereas the information user can be experienced or inexperienced in using search tools or in using computerised media at all, the information broker is always an expert user. He or she can spend time on learning to use a wide variety of tools, and can for example acquire advanced interaction methods for instructing search agents. The information broker can also be provided with advanced visualisation tools such as Spotfire (Ahlberg and Shneiderman 1994) for reviewing information retrieved through search as well as feedback information received from users.
After the collection of information from various sources, the information broker evaluates its relative importance and then chooses whether to include the information as it is, disregard it, summarise it, or perhaps rewrite or illustrate it differently than in the original source. The results from information retrieval are thereby transformed into a few (or just one) annotated hypertext documents that are passed on to the readers. All readers with the same general profile will receive the same annotated hypertext document. The annotations can be used to adapt the information presentation to details in each reader's individual profile. The reading behaviour is monitored, to provide feedback to the editor on the reader's reading pattern, preferences, and (possibly) understanding of the information provided.
The architecture in Figure 1 gives an overall picture of the information flow of the information retrieval task. The knowledge needed to search for, edit, and annotate the information contained in group profiles, generalised from individual reader's profiles. Individual profiles are used to further adapt the information to individual user preferences. To handle this task, several types of tools are needed.
The EdInfo ideas are currently being applied in an Intranet environment at a Swedish company. The information department is responsible for an internal newsletter, published weekly on the Intranet. The newsletter editor gathers her information mostly through a network of contact persons all over the company. She is also provided with information directly from other people who want something to be published in the newsletter. In addition to the central newsletter, some departments and groups have their own newsletters, distributed locally. Some of that information is also published on the Intranet.
The Intranet mainly consists of a series of menu-like web pages in an ordinary hypertext structure that provides access to various information sources. A few examples are: organisational information, manuals and procedure descriptions, a phone book over all employees, forms for all kinds of services such as internal course registration, and finally the internal newsletter(s). There is also a web search facility. The company is currently changing to the second generation of their Intranet solution. The new solution is different from the old mainly in the structuring of the information. Instead of providing an initial menu structure, the navigational pages will consist of two main parts: a selection of links to articles from the newsletter(s), and a selection of navigational links to other information sources and navigational pages. This way, the newsletter articles will be better tied to the topics and navigational levels of the Intranet. There is currently no adaptation to individual users or groups of users in this system.
To enhance the search functionality over newsletter articles, these will not be statically linked to particular pages, but instead retrieved from a database with a web interface. Our first attempt at user adaptivity in the EdInfo project will utilise this functionality to further adapt the retrieval of articles to specific users. When added to the database, each article will be provided with information (such as keywords) to make it easily retrievable. In EdInfo, we will also annotate each article with complementary information to be used for individual adaptation at the reader level. The decision on what information to include will be based on a study of the readers. For example, it is likely that each article will be annotated with its target group (administrative, technical, ...), and whether it mainly concerns a specific group/department, if the article is of a professional or recreational nature etc. It must also be possible to annotate especially important articles as mandatory, to prevent them from being filtered out by the individual adaptation mechanisms.
The personal user profiles will contain the corresponding annotations for individual users, together with phone book information and contextual information (such as, which articles the user has already read). In the first version of the system, we expect that the main method to form this profile will be through explicit user interaction, but the user modelling will also require some simple automatic inference means, in particular to maintain the current context. It is currently not clear whether the computer infrastructure will allow a solution where individual user profiles are stored locally, or whether we will need to use a global user model database.
The adaptations will take the form of an individualisation of the navigational pages, to show only (or sort) article links that are relevant to a particular user. The adaptations will be executed locally, most likely through a Java Applet solution. Note that in the first version of the system, we will only attempt to adapt the links to newsletter articles - the navigational structure will remain unaffected. Technically, the adaptations can easily be extended to cover navigational links as well, by storing information not only about articles but also about all kinds of Intranet pages in the database. However, the design of the Intranet structure needs to be rethought if such adaptations are to be used.
The most challenging task for this first version of the system will be to implement a feedback system to the editor. The individual user logs will be suitably filtered and then sent back to the editor by means of a user logging system developed in a previous project, WebEval. The editor will be able to review the retrieved information by an application of the Spotfire data visualisation tool, in which the editor also can modify and create new user characterisations.
In the first version of the system, we will target solely on internal information, created by a few individuals inside the company. Other individuals work with information from external sources, and publish their information in another part of the Intranet. In the second phase of the project, the two editorial services will need to be connected and integrated. This will put higher demands on developing the full information broker scenario, since the editors working with external sources have higher demands on a full editorial environment, in which information can be retrieved, selected and structured.
We have discussed an approach to handle the search, filtering, and presentation of 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 end users. A similar approach to information brokering is taken in the COBRA project. Our approach differs from COBRA in that we focus on developing feedback mechanisms from information users to information brokers.
We have also described the application of the approach to an Intranet environment. Working with information editing in an Intranet application is simpler than addressing open services on the WWW, for several reasons:
Despite these simplifications, we believe that the EdInfo approach also is applicable to open information services over the WWW. Our aim is to gradually extend the approach to applications in this domain.
The EdInfo project is funded by the Swedish National Board for Industrial and Technical Development, NUTEK.
Bonsall, P. W., and Joint, M. (1991): "Evidence
on Drivers' Reaction to In-Vehicle Route Guidance Advice", Proceedings
of 24th ISATA International Symposium on Automotive Technology and
Automation, Florence, Italy.
Brusilovsky, P. (1996): "Methods and
Techniques of Adaptive Hypermedia", Journal of User Modeling and
User-Adaptive Interaction, UMUAI 6.
COBRA: Cook, R. and Kay, J. (1994): "The Justified
User Model: A Viewable, Explained User Model", Proceedings of
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.
http://www.sics.se/~annika/umuai-final.ps.Z
Höök, K., Rudström, Å., and
Waern, A. (1997): "Edited Adaptive Hypermedia: Combining Human and
Machine Intelligence to Achieve Filtered Information", in notes from
the workshop on flexible hypertext, Hypertext Conference.
http://www.sics.se/~kia/papers/edinfo.html
Maes, P. (1994): "Agents that Reduce Work and
Information Overload", Communications of the ACM,
Vol. 37, No.7,pp. 31-40, 146, ACM Press.
http://pattie.www.media.mit.edu/people/pattie/CACM-94/CACM-94.p1.html
Shneiderman, B. (1987): Designing the
User Interface: Strategies for Effective Human Computer
Interaction, Addison-Wesley.
Spotfire:
http://www.ivee.com/