|Knowledge Broker from University of Edinburgh|
Knowledge Broker fact-file
What's the Problem?
The key to effective knowledge management lies in accessing the right knowledge at the right time. In the context of today's distributed information organisations this may not be a straightforward task. While a company intranet (and, more generally, the internet) would seem to present an ideal medium for delivering such knowledge 'services', the problem arises of just how these services are to be located in such an decentalised, fluid environment.
Towards a Solution
Some sort of brokering service is commonly proposed as a solution to this problem of locating and using disparate knowledge services in a distributed environment. The Knowledge Broker represents an attempt to provide such a service from an AKT, knowledge-centric persepective. Agents offering services advertise them to the broker, while agents with problems to solve pose queries to the broker; the role of the broker, then, is to match one or more of the available services with these queries, where possible, and to suggest these possibilities to the querying agent.
The AKT broker operates at two complementary levels of abstraction:
Describing a broker service (in this case a 'neural network training algorithm' service) in the more abstract form of a knowledge transformation.
When operating at the first of these levels, the broker is able to formulate relatively precise 'chains' of agent interactions (which can often be automated, if the environment allows) to solve precise queries. At the second, more abstract level, the brokered solutions are, as might be expected, more abstract, but which allow progress to be made in situations where precise formulations of queries are, for whatever reason, not possible, or where more general explorations of the space of possible services are wanted.
Constructing a query using the panel
An example of one of the alternative responses suggested by the broker in response to a transformation query: in this case, the desired transformation is achieved by a sequence of 'neural network preprocessing' followed by 'neural-network-training'
This broker is built upon a number of technologies.The brokering mechanisms themselves are written in Prolog, and make use of that language's peculiar backtracking capabilities to select alternatives and construct chains of individual competences into a global solution. Obviously, communications are fundamental to distributed inactions: the broker uses the AKTBus protocol and libraries, consisting of XML layered on conventional HTTP and TCP/IP protocols, to provide basic message-passing capabilities. These messages are further structured through the use of FIPA and RDF/S to allow the unambiguous specification of their components in a manner independent of any particular programming language or paradigm.
Take a Guided Tour
Try a Demonstration
For running the broker: Sicstus Prolog v.3.8+ (including Jasper and Linda modules), Java (with AktBus and Jena libraries), HTTP/TCP/IP connectivity for handling remote broker communications. For querying and advertising services to the broker, Java (with AktBus and Jena libraries) pllus HTTP/TCP/IP connectivity.
Schorlemmer, Dr Marco and Potter, Dr Stephen and Robertson, Dr David and Sleeman, Prof Derek (2002) Formal Knowledge Management in Distributed Environments.