The homogeneous semantic (associative) network built using the neural network 
technology TextAnalyst consists of pairs of associations as they occur 
in the text. The analysis and classification of such associations obtained from 
specific texts show that in fact these associations are classified into classes 
of predicates’ actants. 
The resulting conclusion suggests that using comprehensive linguistic analysis for 
construction of extended predicate structures for simple fragments of sentences, 
it is possible to obtain information on relations 
between key concepts of a homogeneous semantic (associative) networks, 
that is, to move from homogeneous (associative) 
semantic networks to heterogeneous ones. Moreover, these relations 
between key concepts in this case can be marked using automatic analysis. 
The replacement of associative relations in a semantic network by diverse ones, 
in addition to improving the quality of classification and text abstracting, 
will allow for automatic construction of non-homogeneous (heterogeneous) 
semantic networks that so far has been done manually by experts.