Evaluation of a Permutation-Based Evolutionary Framework for Lyndon Factorizations

String factorization is an important tool for partitioning data for parallel processing and other algorithmic techniques often found in the context of big data applications such as bioinformatics or compression. Duval's well-known algorithm uniquely factors a string over an ordered alphabet into Lyndon words, i.e., patterned strings which are strictly smaller than all of their cyclic rotations. While Duval's algorithm produces a pre-determined factorization, modern applications motivate the demand for factorizations with specific properties, e.g., those that minimize the number of factors or consist of factors with similar lengths. In this paper, we consider the problem of finding an alphabet ordering that yields a Lyndon factorization with such properties. We introduce a flexible evolutionary framework and evaluate it on biological sequence data. For the minimization case, we also propose a new problem-specific heuristic, Flexi-Duval, and a problem-specific mutation operator for Lyndon factorization. Our results show that our framework is competitive with Flexi-Duval for minimization and yields high quality and robust solutions for balancing where no problem-specific algorithm is available.