| With the widespread use of databases and their rapid increase in size, there is a need for effective ways to use this data. Current customer relationship management strategies address the needs of the top eighty percent, ignoring many possible customers. Current data mining approaches do not provide the flexibility needed as data grows.; This thesis presents a method to provide probabilistic analysis to very large data sets using conditional probability instead of static rules. Because data tends to be sparse in large data sets, relationships often occur in clusters. By allowing a tradeoff of runtime performance for a loss in precision through clustering Bayesian relationships, we can use the smaller networks to obtain better performance and scalability.; The methods demonstrated could be used as recommendation engines for purchases, ad banner placement and presenting relevant content based on learned customer profiles. |
