On probabilistic models for uncertain sequential pattern mining
conference contributionposted on 28.10.2014, 10:26 by Muhammad Muzammal, Rajeev Raman
We study uncertainty models in sequential pattern mining. We consider situations where there is uncertainty either about a source or an event. We show that both these types of uncertainties could be modelled using probabilistic databases, and give possible-worlds semantics for both. We then describe ”interestingness” criteria based on two notions of frequentness (previously studied for frequent itemset mining) namely expected support [C. Aggarwal et al. KDD’09;Chui et al., PAKDD’07,’08] and probabilistic frequentness [Bernecker et al., KDD’09]. We study the interestingness criteria from a complexity-theoretic perspective, and show that in case of source-level uncertainty, evaluating probabilistic frequentness is #P-complete, and thus no polynomial time algorithms are likely to exist, but evaluate the interestingness predicate in polynomial time in the remaining cases.