A posting from Graham Klyne (Baltimore Technologies) announces the availability of a prototype work-in-progess RDF-driven expert system shell. The RDFExpert software uses Brian McBride's JENA API and parser. Distributed as a single ZIP archive containing a number of Java .jar files, the application provides "a simple expert system shell that uses RDF for all of its input: knowledge base, inference rules and elements of the resolution strategy employed. It supports forward and backward chaining. The tool uses a special vocabulary to build arbitrary n-place predicate facts and rules. Other RDF statements are interpreted as binary facts (i.e., predicates with constant arguments). There is also a representation for n-place predicates that generalizes the normal RDF representation of binary facts." The RDFExpert web site provides a manual, sample test cases, and an overview document which outlines the influences, capabilities and future directions of a research project entitled the RDFExpert' undertaken at the strategic research department of Baltimore Technologies. Graham writes: "Craig Pugsley has been working on an experimental expert system shell that uses RDF for all of its input data (knowledge base, rules and 'control'). We can now use it to read arbitrary RDF from the web and perform inferences on that data. We have been exercising this capability using RDFweb/webwho data. A simple example query we have run is to list RDFweb people with a common interest... It is very much an experimental/prototype piece of software, and all the usual caveats apply about being provided as-is, without warranty, etc. Further developments under consideration for the tool include support for an inference rule format compatible with RuleML work; forward chaining from a designated set of facts; rules containing variable predicate names; extension of resolution strategy component to provide greater control over fact resolution process."

A manual is available for the RDFExpert and associated applications: "All screenshots shown are from the RDFExpert running within Mac OSX. Included is a short description and usage guide to the associated application RDF2Text' which converts RDF-encoded knowledge graphs to predicate fact representations. The final section outlines the schema... The RDFExpert works by repeating a cycle of match-decide-act. When the RDFExpert is asked to infer for the first time, it's first point of call is the Resolution Strategy definition. The Resolution Strategy contains all the information needed to tell the inference engine what rules to consider, and what inference method to use to consider these rules. It is the job of the Resolution Strategy, therefore, to guide the direction of inference to obtain (infer) the most specific knowledge to a given goal in an attempt to resolve the goal. In this version of the inference engine, there are two resolution mode that can be used to resolve a goal. These methods of inference are called backward and forward chaining. Backward chaining is the process of trying to match the goal fact against the conclusion element of rules in the knowledge base -- moving backwards from the goal through the rules. Forward chaining works by trying to match rules in the knowledge base against existing facts in the knowledge base -- thereby moving forward towards the goal fact through the rules. If rules are matched, a sub goal to be considered later is created in the inference core. Sub-goals are then placed on an ordered list called an agenda' which dictates the order in which rules are activated..."