This paper describes the initial steps of a project with the ambitious goal of developing a distributed non-supervised system that will allow the incremental structuring of knowledge. The main motivation is to enable the creation of high quality course materials as a result of the interaction of the students with the current materials. The name of the tool - KnowCat - stands for "Knowledge Catalyser": the main goal of the tool is to catalyse the crystallisation of knowledge as the result of the interaction of people working with it.
The application area of our proposed system will be education, more precisely, the area of university courses. Learning at university happens as an interaction of several activities. Some of these activities require the study and comprehension of established pieces of knowledge. Other activities are related with the use of this knowledge to solve practical problems. Finally, some of the activities require the student to critically review an area of knowledge and to elaborate his/her own view on it (for example, see [Barros et al. 1997]). We pretend to enhance the performance of these later activities by means of a computer supported collaborative work (CSCW) environment [Witaker 1996].
KnowCat will enable the participation of several students in the process of reviewing and elaborating the knowledge. Therefore, the tool has a component related with traditional CSCW work. The innovative aspect of KnowCat, and in fact the real goal of the tool, is to motivate the students in such a way that the result of the interaction of the students with the course materials is an improved version of these materials. These improved materials will be the input for the students following, that in this way will take advantage of the efforts of previous students and, in turn, will make their own contributions. We expect that the materials will eventually evolve to the optimal with respect to educational use.
As stated in the introduction, the goal of KnowCat is to allow the crystallisation of knowledge as the result of the interaction of students working on that knowledge. In this section we are going to spell out some of the characteristics that KnowCat will exhibit.
In first place, the knowledge will evolve in a non-supervised way. This is a very hard requirement that has been included, among other reasons, to ensure that the system will be scalable. One obvious way of achieving knowledge crystallisation is by having an "editor" (probably the professor of the course) doing a "reification" of the materials produced by the students [Lewis 1995]. In fact, this is the standard solution to this problem today. However, we believe that this solution is not very practical. This editor role is too demanding, and to achieve results of quality you need people dedicated full time to this work. One of the initial motivations of KnowCat was to avoid the need of such centralised "editor" role: knowledge would evolve as the result of the interaction of several people, with no "supervisor" on duty.
In second place, the system will be scalable. We are thinking in a system that will be initially used for creating and maintaining small corpuses of knowledge related with university courses, but that tomorrow may be used to articulate all these corpuses in a larger framework, or may be used in other environments, such as to manage the knowledge in a company. Scalability has a profound impact in implementation issues.
In third place, the system will be able to work with knowledge in "evolution". In each moment, there will be many pieces of knowledge in the system, each in a different status of crystallisation. Knowledge recently added is very fluid. Knowledge that has been around for a long time and that has qualified as useful is very stable (crystallised). Of course, the mechanisms for crystallising (and sometimes de-crystallising) knowledge are a central issue in our research.
Finally, the basic knowledge contributions themselves are not part of KnowCat. They are plain WWW (World Wide Web) pages located somewhere in the Web. KnowCat essentially creates a meta-structure on top of the current WWW, classifying and organising some of the Web pages in the form of a "knowledge tree".
We are currently implementing the first version of the tool, which will be used during the academic year 1998/99 in three university courses at the Universidad Autónoma de Madrid. Validation of our initial ideas and further refinement will be obtained from this prototype experimentation. The rest of the paper describes some implementation details of the prototype.
Knowledge contributions are not part of KnowCat. That is, generating new knowledge (for example, a paper on the foundations of probabilities) is a process that is not managed by KnowCat. However, KnowCat presumes that any contribution is a WWW (World Wide Web) document located somewhere in the Web. What KnowCat maintains is a meta-structure of (some) of the Web documents. This structure has two main goals: to classify the documents in a hierarchy and to exhibit the grade of crystallisation of each one. This meta-structure is essentially a classification tree. Each node represents a "topic", that is, an entry to the encyclopaedia. For each node there are two associated items: a set of mutually alternative descriptions of the topic (that is, a set of addresses of WWW documents with such descriptions) and a set of mutually alternative further refinements of the topic. A refinement of the topic is a list of other nodes of KnowCat. In each set there is a dominant element: the (current) best description and the (current) best refinement of the topic. Any other element is an alternative to this "best" candidate and is competing to occupy its place. Elements that are not successful enough will be eventually discarded (the link, not the document).
In some aspects, KnowCat is equivalent to some searching engines, which maintain a meta-structure of the Web and facilitate the search for useful information [Asnicar & Tasso 1997]. However, there are three main differences among search engines and KnowCat. The first and more important is that the meta-structure in KnowCat is the result of the asynchronous, non-supervised and distributed interaction among several users and the information. The meta-structure is the result of human work, but no editor is in charge. Secondly, the meta-structure in KnowCat represents directly a hierarchical ontology of the knowledge. Finally, KnowCat is implemented in a distributed way: each node of the hierarchy may reside in any KnowCat server. KnowCat servers, in fact, are essentially WWW servers.
Perhaps it is interesting to note in this moment that authors may submit two types of "contributions" to KnowCat. The first type is a "content" contribution: for a existing concept in the KnowCat hierarchy the author submits the WWW address of a document that is proposed as a description of the topic associated with the node. This contribution will begin to compete for crystallisation against the other alternative descriptions. The second type of contribution is a "structure" contribution: the author submits an alternative refinement for the topic. This type of contribution also competes with other alternative refinements.
When a user (a student) wants to use KnowCat, he/she connects to the adequate KnowCat root page. It is not required that all the KnowCat hierarchies are connected in a global knowledge tree (although this would be a desirable long-term goal). However it is expected that there will be a KnowCat root topic for each reasonably broad area of knowledge. For example, in the three experiments to be conducted, the root topics will be "operating systems" for two of them and "reasoning under uncertainty" for the third. The student will navigate through the hierarchy, eventually stepping out to read some of the contents associated with them. From the use of the elements (contents and structure elements) and from explicit voting on their quality, the crystallisation will occur.
Crystallisation of contents and structure is, of course, a central aspect of the system. The initial proposal is based in the following ideas. Each element in KnowCat (both contents and structure) will have an associated crystallisation grade. This grade is made from the number of accesses to the element (in relation with the total number of accesses to any of the elements related with the same topic), from explicit users votes and from the amount of time that the element has already survived. The crystallisation grade of each element will evolve, according to its perceived utility as judged by the user community. The best contribution will eventually achieve a stable (crystallised) status, while the competing ones will disappear. However, new alternatives may always be submitted, and competition is never final for any piece of knowledge. Of course, the more crystallised, the more difficult is to compete with a given piece of knowledge.
Note: In the final paper a more detailed description of the crystallisation process will be shown, as well as some details in the design of the experiment to be conducted during the next academic year and some thoughts on future work to be done.
[Asnicar & Tasso 1997] Asnicar F. A. & Tasso C. (1997)."ifWeb: a Prototype of User Model-Based Intelligent Agent for Document Filtering and Navigation in the World Wide Web". In the proceedings of the workshop "Adaptive Systems and User Modeling on the World Wide Web", Sixth International Conference on User Modeling, Chia Laguna, Sardinia, 2-5 June 1997.
[Barros et al. 1997] Barros, B., M. Rodríguez-Artacho and F. Verdejo. "Towards a model of Collaborative Support for Distance Learners to Perform Joint Tasks", in the Proceedings of the 1997 IFIP TC3 Joint Working Conference (The Virtual Campus: Trends for Higher Education and Training), Madrid, November 1997, pag. 123-136.
[Lewis 1995] Lewis, R. (1995). Just-in-Time Open Learning - a DELTA project: concepts, perspective and futures. Neurope Lab - Occasional Paper: NL/8/95.
[Whitaker 1996] Whitaker, R., CSCW and Groupware: Overview, Definitions & Distinctions, http://www.informatik.umu.se/~rwhit/CSCW.html.