(eng): Like Bayesian networks, compositional models may also be used for data mining. Nevertheless, one can find several reasons why to prefer compositional models for this purpose. Perhaps the most important is the fact that compositional models are assembled from low-dimensional (unconditional) distributions so that computationally advantageous formulas are known for information theoretic characteristics computation. The other reason is that a decomposition is a natural way of complex tasks simplification. Therefore, the inverse process of composition is easily understandable for specialists from many fields of applications regardless of their level of mathematical education.