A Framework for Frequent Sequence Mining under Generalized Regular Expression Constraints

This paper provides a framework for the extraction of frequent sequences satisfying a given regular expression (RE) constraint. We take advantage of the information contained in the hierarchical representation of an RE by abstract syntax trees (AST). Interestingly, pruning can be based on the anti-monotonicity of the minimal frequency constraint, but also on the RE constraint, even though this latter is generally not anti-monotonic. The AST representation enables to examine the decomposition the RE and to choose dynamically an adequate extraction method according to the local selectivity of the sub REs. Our algorithm, RE-Hackle, explores only the candidate space spanned over the regular expression, and prunes it at each level. Due to the dynamic choice of the exploration method, this algorithm surpasses its predecessors. We provide an experimental validation on both synthetic data and a real genomic sequence database. Furthermore, we show how this framework can be extended to regular expressions with variables providing context-sensitive specification of the desired sequences.