The amount of information on the Internet has been increasing with the accelerating speed. The problem we are facing is how to ensure the quality of information within the enormous amount of information. The movement of Semantic Web is the straight answer to the problem, i.e., providing the knowledge level markup to documents which are understandable intelligently both by human and machines. Since markup tags are based on ontologies, shared understanding is ensured. The mechanism is rational, but then the problem is shifted to “how to provide ontologies” or “how to construct semantic descriptions”?. Simply providing some “good” ontologies is not sufficient, because our semantics in the real world is dynamic in nature. Our semantics is spread from surely shared one to very domain dependent or private one. Meaning sometimes comes from the domain-dependent semantics and becomes public. Meaning in public is sometimes transformed into the domain-specific one. Meaning is thus dynamic, so we should support such dynamics of meaning to realize truly useful knowledge-level descriptions.
Towards support of the dynamics of semantics, we started a project called “Collection and integration of concept systems” (CICSS). The goal is to provide an infrastructure that can be used to generate new conceptualization for new requirements for conceptualization by mixing and combining existing concept systems. To realize the goal, the mission of the project is very simple, i.e., just collecting various concept systems in various fields or domains and relating them loosely.
One or more concept systems usually exist in each field / domain ether implicitly or explicitly. They are very good sources of knowledge for that domain. All concept systems show their view of categorization at least. Since the most primary nature of knowledge is categorization. Dealing with concept systems as categorization is expected to explicate the most basic level of knowledge. We propose concept systems repository as the workspace for the above research. The basic functions of concept systems repository (CSR) are as follows (Fig. 1): - Import concept systems - Relate concept systems - Retrieve concepts or concept systems The strategies to create CSR are (1) treating concept systems as a whole not as individual concepts, and (2) Emphasizing variety of concept systems not hiding it.
The tasks to realize CSR are twofold. One is modeling concept systems to treat concept systems in a system, and the other is the algorithm to find relationship among them. We will consider the latter task more in the next section.
Finding relationship among different concept systems is a very difficult problem. We do not have the exact answer yet, but show some preliminary consideration and efforts for it.
We adopt the instance-based or extension-based approach for discovery of relationship among concept systems. Definition-based or intension-based approach is not applicable in principle. If definitions from two concept systems are comparative, it tells us that they are essentially the same conceptualization. If they are formed by the different conceptualizations, definitions are not comparative. Since the instance-based approach does not care definition, it is suitable for our purpose. Of course the instance-based has other problems. One is how to identify instances, i.e., even definitions of instances can be different. The other problem is how to approximate instances, i.e., since instances can be enormous in nature, we will need some methods to say “almost the same”.
We have two trials based on this line. Fig 2: Results of mapping of categories between two directories (Literature case)
One is a method to align different Internet directories like YAHOO!,
Lycos, and so on. The Internet directories are not strictly defined
but very elaborative and practical concept systems for the Internet.
In this work, we proposed the method to generate mapping rules
from categories in one Internet directory to the other by the
statistical method[
The other work is to find human relationship through WWW
bookmarks. WWW bookmarks are results of implicit or explicit
efforts to represent personal views for the Internet [
As the evaluation as recommender systems is good enough. Topics found by the system were apparently more acceptable than pages themselves.
We described a project called collection and integration of concept systems (CICSs). The goal is to provide a repository of concept systems that can be used to yield new conceptualization. In some sense, it is similar to Cyc, but Unlike Cyc, we do not create but collect knowledge.
Our preliminary consideration leads to adopt the instance-based approach for discovery among concept systems, because concepts under different conceptualization cannot be compared directly.
Fig 2: Mapping among WWW bookmarks