Catalogue of Artificial Intelligence Techniques
Case Retrieval Nets
Aliases: CRN, CRNs
Keywords: case, case-based, cbr, crn, crns, data representation, knowledge, memory, net, nets, network, networks, reasoning, retrieval
Author(s): Michael Ichnowski
Case Retrieval Nets (CRNs) are one of the knowledge representation models used in Case-Based Reasoning (CBR). They were created to mimic the way humans search through their memory in order to recollect similar situations that happened before and could offer a ready solution or a clue to a current problem. Apparently, people don't search through each and every case observed in the past but are able to choose the ones that seem to be relevant to the situation.
CRNs, as the the name suggests, organise information in a network. The basic elements constituting the network are the so-called Information Entities (IEs) made up of a pair of an Attribute and Value. Each IE is stored as a node, possibly connected to other nodes with a mutual similarity arc, based on an observed relevance. The other kind of nodes present are the particular Cases, each of which may be reached by multiple IEs by means of the relevance arcs.
When searching through the network, we start from the IEs presented in the query and work our way through to the neighbouring IEs using the similarity arcs. Whenever we find a relevance arc, we collect the Case Node information referred by it. Each arc is marked by weight indicating the degree of similarity.
CRNs prove to be much more efficient and flexible than other techniques of the same purpose. Some of their key characteristics are:
- Working with partly specified queries without the loss of efficiency.
- Completing the Cases from the given fragmented information.
- Ability to change the degree of similarity and relevance at runtime by changing the weight values associated with the corresponding arcs.
- Queries can contain attributes not relevant to the searched case and still return proper results.
- New nodes can be added by simply connecting them to the relevant and similar nodes by means of proper arcs. The added nodes can contain totally new attributes.
- Amílcar Cardoso, Gaël Dias, Carlos Bento, Lisboa Bento Bento, Progress in Artificial Intelligence: 12th Portuguese Conference on Artificial Intelligence, EPIA, Springer-Verlag, 2005, pp.56-58.