For the SIGART EIS, an ontology is being developed which richly represents the concepts relating to the information being stored. The domain of the EIS is currently limited to news and announcements of interest to people in the AI field, such as conference CFPs and programs, journal calls, job announcements, etc. This limited domain is ideal for testing these ideas in a working environment.
The ontology itself begins, as shown in Figure 1, with the familiar universal concept thing at the root of the taxonomy, followed by the subconcepts represented-thing and system-thing, which respectively denote concepts to be represented in the domain model and concepts known internally to the computer (such as numbers, characters, etc.). Below represented-thing are domain-thing and ontological-thing. The concept domain-thing subsumes all the top-level domain-dependent concepts (in other words, the ground concepts for each separate domain), and ontological-thing subsumes all concepts for ontological categorization, which currently is limited to actions and objects.
There were two domains that needed to be represented for the purposes of this model and they are shown in Figure 2. The top-level concepts for these domains appear below domain-thing in the taxonomy, UNIX-thing and EIS-thing. Below UNIX-thing are such concepts as UNIX-file, UNIX- command, UNIX-process, etc.
The EIS-thing concept is further specialized into three objects shown in Figure 3. They are, briefly, information-item (a piece of information, such as a Call for Papers), information-view (a way of viewing an information item, such as video or text), and information-topic (concepts about which information is stored, such as a conference. A user will typically not wish to retrieve the concept corresponding to the conference, but the items of information about the conference).
The concepts information-topic and information-item can, at first glance, be confusing, but the need for both in a system like the EIS is clear. Consider the concepts in Figure 4. Here an instance of person (a specialization of information-topic) is shown called Chris. This instance has an item of information associated with it, chris-resume, which is an instance of resume, a specialization of information-item. There is a lot that needs to be known about chris-resume that has nothing to do with Chris, but has to do with chris-resume's role as a piece of information that will be retrieved by users of the system. For example, as shown in Figure 5, there is a Postscript and an ASCII text view of the resume, and each is stored in a separate UNIX file - clearly this is not information that should be part of the object Chris.
This example shows the need for representing information items, especially to anyone who has tried to get information from an electronic source (such as FTP, gopher, WAIS, etc.) and is familiar with the notion of retrieving files containing such things as a resume in postscript format. Information topics, however, offer a new dimension for information retrieval of this nature by representing as separate objects the people, conferences, etc. that the information being retrieved is about. Having this information available makes it possible to tie information items together in a way that is not possible in the network information retrieval systems mentioned.
For example, a user of the EIS who is interested in AI sends a query for the programs for all AI conferences in Texas in 1994, and is returned one of the views of program-23, which is the program for the conference CAIA-94 (the default view is the ascii view). The representation for the knowledge associated with this query is shown in Figure 6. The user is interested in some of the titles in the program, and decides to order the proceedings.
At this point the user has already gone far beyond the capabilities of any existing information service, with the possible exception of an on-line library catalog system. Searching for information as specific as ``AI conferences in Texas in 1994'' is not generally supported. The EIS domain model goes still further however, since the representation includes the relationship between a conference, the proceedings for that conference, and the publisher of the proceedings. This is part of the intelligence that, from the user's perspective, the system exhibits.
Having decided to order the proceedings, the user now sends a query to the system for information about the publisher of the proceedings for CAIA-94, and is returned one of the views of the information item description-13, which is the description of IEEE-press.
This example shows that more than files containing information like resumes or conference programs can be stored and retrieved in the EIS. The domain model includes an understanding of all the objects in the domain and their inter-relationships, and adding this to the information service makes it easier for users to find the information they need.