Mining Interesting Itemsets using Submodular Optimization

We propose a novel technique to retrieve itemsets that best explain a transaction database by leveraging a simple probabilistic model. Our approach is the first to infer such interesting itemsets directly from the transaction database using submodular function optimization and in so doing avoids many of the pitfalls commonly present in frequent itemset mining algorithms. Our proposed approach is theoretically simple, straightforward to implement, trivially parallelizable and exhibits good performance as we demonstrate on both synthetic and real-world examples.