Ann Wrightson, February 2001
There has been a lot of excitement and speculation concerning the potential role of Topic Maps in Knowledge Management, and especially concerning their potential for Knowledge Representation (KR). This whitepaper takes a critical yet hopeful view of this potential, looking briefly at some classic areas of KR to see where Topic Maps can take their rightful place as a new and complementary technology.
Over the last few months, I have participated in several very interesting discussions about the possible role(s) of Topic Maps in KR. Some participants were convinced that Topic Maps are a KR revolution for the Web; some were very sceptical about their usefulness, citing the enormous performance and implementation problems encountered by many KR technologies when they tackled industry-scale problems; and there were many other views and experiences too. So how can we find a way through all this to a fair and realistic view of the potential for Topic Maps in KR?
I believe that the key characteristics of Topic Maps for KR are:
The first has an appealing similarity to Semantic Networks, a simple KR concept which is very popular in textbooks and first courses in AI and Knowledge Based Systems. However, effective Semantic Network based applications and tools need much more than the raw network: simple associative networks have been found to be lacking in representative power (it's difficult to use them to say just what you mean) and also difficult to search effectively when they get large. So although Semantic Networks are a great idea, for effective industry-scale applications, the basic idea needs to be refined and combined with other KR techniques. There remains the intriguing possibility that a Topic Map could help some of these applications "travel in Webspace", but the immediate appeal of the raw Topic Map network is unfortunately misleading.
However, the other characteristics listed above are much more promising. Each of them is explored in turn below, in relation to a well established branch of KR which could be enriched and enabled by Topic Maps within the Semantic Web.
CBR (Case Based Reasoning - see http://www.ai-cbr.org for many resources including those quoted here) has enjoyed more business success than most KR techniques. CBR is based on the idea of learning and acting from experience, with known "cases" serving as patterns for reponses to similar new situations. CBR is also highly relevant in corporate Knowledge Management - the value of being able to retain and learn from experience is evident in the popularity of slogans such as "learning organization" and "corporate memory".
So where do Topic Maps come in?
CBR has four main components (Aamodt & Plaza 1994):
Topic Maps are potentially most helpful for (1) - their ability to enrich legacy resources with metadata, and, via the query language TMQL (under development), to enable those resources to be identified and retrieved according to topic templates or profiles, looks very well suited to this application. However, no functionality, not even a good Topic Map, is free! In order to enable a Topic Map to fulfil this role, you will need to design the application carefully, working out in detail: how the Topic Map "clothes" the legacy resources; how the metadata is generated (eg from tables in the legacy data); and how the Topic Map queries can be optimized to yield good user responses.
Components (2) and (3) above are less likely to be directly related to a Topic Map - but with step (4), a Topic Map is again really useful - the new "case" can be stored in a current storage area, and also linked into the Topic Map so that it will be readily available in future alongside the legacy material.
There is also another potential benefit of using Topic Maps to integrate new CBR applications: because Topic Maps are inherently part of the new generation of Semantic Web technology, enabling such a CBR solution to work across sites and across platforms is likely to be significantly easier using Topic Maps than using platform-specific technologies - especially as vendors of Web-aware repositories start offering Topic Map interfaces as standard on their products.
Ontologies are emerging as a key aspect of information management in many areas, from the interchange of engineering data to corporate knowledge management.
So what's an ontology? - briefly, an ontology is a way of describing a shared common understanding, eg about the kinds of objects and relationships which are being talked about, so that dependable communication can happen between people and application systems. An informal example is the terminology of a subject (eg we all understand the terms "relational database" & "software"); a more formal example is a list of controlled terms, eg a thesaurus used as a subject index.
So where do Topic Maps come in?
Topic Maps are very well suited to representing ontologies (this is by design not by accident!). Because of the key role of ontologies in many real-world KR applications, the ability of Topic Maps to link resources anywhere in the Semantic Web, and then organize these resources according to a single ontology, will make Topic Maps a key component of the new generation of Web-aware knowledge management solutions.
In addition, the growing repertoire of techniques for simplifying, merging and interrelating ontologies can be used to combine or articulate Topic Maps representing different ontologies, thus enabling disparate sets of information resources to be used together in a controlled and scalable way.
This, if anything, is the key capability for realizing the vision of the Semantic Web. There is no realistic hope of globally classifying all concepts, terms and relationships; we need to be able to manage and interrelate our ontologies project by project, domain by domain, so that scalability is achieved without either runaway complexity or over-simplification.
There is one current development in particular which is of interest in this context. Amongst the Ontopia demos at XML 2000 is a simple proof-of-concept for integrating Prolog-like rules with a Topic Map. A small beginning - but because of the inherent simplicity of Prolog itself, a very significant one.
Copyright Ontopia AS, 2000