Knowledge Representation and Advanced Reasoning Engines

Submitted by Peter Gabel on Fri, 06/04/2010 - 9:01pm

Arity matches the problem to be solved to the style of knowledge representation and reasoning used. There is a wide range of choices to make in this process.

Relational databases are great for collecting and processing transactional data. The records in the relational database are then queried to match some set of criteria. The Resource Description Format (RDF) “triple” is a style of knowledge encoding that attempts to extend the database paradigm of storing facts and querying against it. RDF represents an important but incomplete special case of knowledge representation and reasoning.

However, there is a lot of knowledge that does not encode well in the RDF style and a lot of things that you want to do for reasoning that do not fit the querying paradigm. Description logics provide the opportunity to model the world faithfully and more completely so that reasoning over them becomes practical. Typically, these descriptions are concept-based. There are currently two competing views about the representation of concepts. One view is that a concept such as “people” includes, by extension, the set of all people as instances of that concept. The other view is that the concept is a definition of who/what belongs to the set of “people.” Definitions and descriptions create a more complete picture and can include relationships between items such as “owns” or “part of” etc. Axioms that encode knowledge can be based on the descriptions and relationships represented in this more complex model. Reasoning in the form of classification, similarity, and analogy become feasible to utilize in this context.

Arity has developed an advanced reasoning engine based on OWL 1.1’s EL ++ fragment. This makes a unique contribution by analyzing implicit knowledge and making it explicit.

In Arity’s products, these reasoning tools improve investigations by:

  • Suggesting alternative areas for research using “reasoning by analogy” across similar entities (you may want to consider B as you are looking at A and B is similar to A);
  • Suggesting new items to consider using “reasoning by conclusion” (you may want to consider X as you have already considered A and X is very likely a conclusion of A);
  • Providing the exploration of concepts by having them linked via explicit relationships and associations rather than relying on search; and
  • Indicating the strength of inferences through confidence factors.