L.S. Kuravsky, E.G. Ivanova

Trained multifactor Markov networks that make it possible to represent subtle features of development of psychological characteristics and study their interaction are under consideration. These structures may be treated as a special kind of neural networks. Decomposition technique simpliying significantly both the model identification and calculation of state probability functions is presented. Following this technique, an initial system under study is decomposed into certain derived subsystems of fewer dimensions, with the states representing state sets of the former system. Each subsystem is identified separately. As a result, obtained probability functions of the state sets may be recalculated into ones of the initial system. Shown examples of 2-D and 3-D multifactor Markov networks representing development and interaction of phonological awareness characteristics demonstrate new types of characteristics suitable for further psychological analysis. The concept in question yields new information for analysis that cannot be obtained with the aid of other approaches.