1988 Volume 3 Issue 4 Pages 503-510
We have been studying a knowledge based approach to Information retrieval and have implemented an experimental knowledge based documentation browser called KBrows. KBrows incorporates documentation databases, knowledge bases and an inference engine. The KBrows knowledge base is organized in the form of semantic networks : keywords in documentation are denoted by nodes, and each link expresses a relationship between two keywords. A Portion of the original document, which describes the relationship, is registered to the link. The automatic exploration is achieved by searching meaningful paths on semantic networks and gathering portions of the document along the paths. In particular, we focus attention on two major problems in building and exploring semantic networks on documents ; stepwise knowledge base refinement, and exploring large semantic networks. It is usually not necessary to construct a precise knowledge base from the very beginning. A more realistic approach is to construct an approximate knowledge base at first, and then refine it as the necessity arises. We have studied this stepwise refinement approach by embodying it in KBrows. To support stepwise refinement, KBrows allows users to represent knowledge in various levels of refinement, and enables characteristics features of relations to be specified incrementally. Efficient exploration of large semantic networks is achived by using the KBrows production system, in which new mechanisms are introduced for focusing on exploration areas. We are in the process of constructing a knowledge base for "Common Lisp : The Language" (CLtL). Relationships between keywords are initially extracted from documentation based on both text structure and Positions of keywords. Refinement of the network is currently being done interactively. Future issues obtained from this experience are also discussed.