A Method for Quantifying the Importance of Facts, Rules and Hypotheses

T. J. Parsons
1987 Procedings of the Alvey Vision Conference 1987  
A labelled digraph is used as a model of a simple database, nodes representing facts (or classes of facts) and arcs the relationships between these facts. An expression for the number of microstates in which such a data structure may exist is derived and used to calculate a measure of intrinsic entropy. This measure is fundamentally related to the information content of the structure and its change, on adding or subtracting an item of information from the database, may be used to associate a
more » ... ue called Importance with the item. An algorithm called the " Database Monitor Program (DMP) " is introduced. Its function is to guarantee that the digraph database exists in a ' least complex state' by replacing relationships between nodes by relationships pertaining to classes of nodes, but with the constraint that the information content of the structure remains unchanged. Once in this minimal condition, the Importance of an item is denned in terms of the change it induces on the minimum entropy state. Like measures of entropy, the Importance of an item is a relative concept in that its value depends on the context of the database. It is argued that this is an intuitively appealing measure in that a system is only able to judge the importance of an item in the light of existing knowledge. Two additional concepts termed confidence and significance are introduced and used to assess the formation of 'class concepts' within the database. The use of these three measures for conflict resolution is also discussed. Finally, an example system developed within the Poplog environment is presented and extensions of the work are discussed.
doi:10.5244/c.1.7 dblp:conf/bmvc/Parsons87 fatcat:oh43gtnv2jbdfmknor42pphg64