Using neural networks for recognizing matching among terms

Some experiments have been carried out in order to identify a class of neural networks that would receive as inputs (n+1) terms {T_l,…T_n.T} and would suggest which T_is are matched by a subterm of T, with varying n. A particular net was found that is able to solve the problem for n=1 with a good degree of accuracy. It is shown that the performance of the net is relatively independent of the size of the training set while it is strongly related to the length of the term T. This net was then used as building block for larger nets that were applied to solve the problem for higher values of n. Results of testing such nets are presented.