Despite their reputation in IR circles as cumbersome, inefficient and
suitable only for small databases, at least one IR researcher has used
knowledge-based indices successfully [FHL+91]. Fuhr et al's
AIR/X performs automatic indexing of documents using terms (descriptors)
from a restricted vocabulary. Probabilistic
classification determines indexing weights for each descriptor
using rule-based inference.
After building a system similar to AIR/X, Jacobs [Jac93] determined that ``the combination of statistical analysis and natural language based categorization is considerably better than either alone.'' His paper describes an automated set of statistical methods for pattern acquisition that operate inside a knowledge-based approach for news categorization (an area closely related to document classification and other information retrieval tasks).
Both of these systems attempt to automate the process of generating index terms. KEYNET, by contrast, is not directly concerned with how the index terms are obtained but rather in the data structures and indexing techniques that would make use of the index terms.
Several distinct families of databases for semantic networks have been developed. Such databases are often called knowledge-base systems. Some of the best known of these are: Conceptual Dependency, ECO, KL-ONE, NETL, Preference Semantics, PSN and SNePs (see [Leh92]). All of these support link types, frame systems and so on, but few if any explicitly concern themselves with performance measures familiar in traditional database work, such as minimizing the number of disk accesses required to retrieve complex structures (i.e., graphs assembled from multiple frames and their typed relations). Contemporary knowledge-base systems traverse semantic networks one frame/relation at a time. Unless the knowledge-base fits entirely in the main memory of a single processor, these traversals result in large amounts of virtual memory paging.
In [Lev91], Levinson describes a technique for pattern-oriented (graph) retrieval. Levinson's approach discovers common structures among graphs in a database, and then uses pattern associativity to reduce the required number of tests for subgraph isomorphism. A separate paper, [Lev92], compares techniques for subgraph isomorphism testing suitable for pattern-oriented retrieval. The baseline retrieval method described in [Lev92] considers a flat set of N networks with no pattern associativity information, in which each query requires N isomorphism tests. The first improvement indexes commonly occurring substructures in the graphs, and tests only a subset of the entire database. A second improvement creates a multilevel index of subgraph relationships between successive levels of substructures and the domain of graphs in the database (establishes a partial order). A final elaboration applies multilevel indexing to connectivity and label information required during subgraph isomorphism testing (via the refinement method), rather than to the graph substructures themselves. This produces a tree of ``node descriptors'' in order of increasing specificity, with actual index pointers at its leaves.
In KEYNET we adapt subgraph isomorphism techniques to determine the relevance of documents that were previously annotated with keynets. Our approach compiles (fragments of) search paths through a tree of node descriptors into a hash table, reducing lookup computation (although increasing database construction overhead).
The KNOWIT system of Sølvberg, Nordbø and Aamodt [SNA92] embeds a semantic network in a front-end query refinement system. Queries to the ESA-QUEST bibliographic database are expanded according to a semantic model that describes the meaning of a concept entirely in terms of its relations to other concepts in the model. The KNOWIT system is a front-end to a traditional IR system, whereas KEYNET uses a different kind of search strategy.
Chu and Chen [CC92] explain that cooperative query answering (CQA) substitutes retrieval terms to and from some abstract object representation (such as a type abstraction hierarchy). Separate frameworks can be used for grouping ``data'' (collections of objects with the same type and many common properties) and ``knowledge'' (relationships between objects of different types in specific problem domains), for example data by (type) class and knowledge by subject. (The ``pattern instances'' that link data and knowledge are mildly reminiscent of semantic network nodes.)
One example of CQA is ``Neighborhood query answering'' (NQA) which first generalizes query terms and then re-specializes them into a collection of ``nearby concepts'' in a compound query. Neighborhood specifiers such as ``morning'', ``afternoon'' or ``cross-country flight'' substitute for specific times or airline names. Chu and Chen present a formal system of rewriting rules and nearness measures for query relaxation in NQA.
Another example of CQA is ``Associative query answering'' (AQA) which traces behaviour dependencies among ``cooperating objects'' under a given subject. Virtual ``pattern instances'' with restricted scope and behaviour provide many-to-many linkages between data objects and knowledge subjects. Knowledge hierarchies based on generalization, composition or abstraction are used to support deductive reasoning that obtains information relevant to query construction but not explicitly contained in a user's request, using a logic-based rule language and inference engine, flexible goals, object contexts, and so on.
In general, CQA is a front-end to a traditional database management system. Moreover, the user is required to specify the degree of relaxation explicitly; it is not done automatically.
The EDS TemplateFiller system [SMHC93] applies Message Understanding (MUC) text-filtering techniques to the generation of knowledge frames for one or a few specific subject areas from entire texts (computer product announcements). TemplateFiller fills in slots for frames that exist in a predefined schema of templates, ignoring subjects that are not in the schema.
There are many other MUC-style systems; in fact, there is an annual competition among them. Such a system could automate the construction of the keynets for a collection of specialized textual documents. In fact, we are building a system of this kind for biological research papers [BFH+93].