Use of sampling and Ant Colony Optimization for predicting support in Recommender System

Recommender Systems based on Association rule mining require a fair estimate of the support required for mining frequent itemsets and thereby generating association rules. Also for large datasets passes over the database are an expensive option so we have sampled the datasets and run frequent itemset generation algorithms on these random samples. The databases are characterised by their support and confidence values, number of frequent itemsets and number of association rules generated for these specific values, the cardinality of the frequent itemsets generated and the number of items in the datasets. We have used the system with sampling as well as generation of rules based on conditional probability using Ant Colony Optimization(ACO). We have used these methods to predict support for a stock market recommender system but it can be easily extended to other recommender systems as well.