Belief revision as database update

Belief revision is a fundamental process that underlies numerous forms of intelligent behaviour. Intelligent information systems must be adept at modifying their beliefs in a rational, coherent and consistent fashion. We show that iterated belief revision strategies can be implemented using standard relational database technology. We illustrate the main ideas using Spohn's (1988) notion of conditionalization. For the purpose of motivation we focus our attention on a problem in market research, namely modeling changes to consumer preferences. The key idea is that possible product profiles can be represented as tuples in a relational database, and the consumer's preference for products is captured using an ordinal ranking over the tuples. Using this representation iterated belief revision strategies can be implemented simply and efficiently. In particular, belief revision is performed using database transactions that modify this ranking.