Model reference fuzzy-neuro text matching system
In this paper is presented Model Reference Fuzzy-Neuro Text Matching System. By using user-friendly interface in the form of if-then rules, syntax and semantics rule base, and English word base, the input text has been transformed to an array of potentially important terms (entities). Entities, that are characterized with their names and attributes that have been extracted from the text, are inputs into the Fuzzy Inference System (FIS). The output from FIS is “term importance” for given text which is regarded as the coordinate of input text in term space. In order to simplify tuning of parameters of the FIS, FIS is transformed to neural network. Neural network parameters are adjusted offline using some reference knowledge that has been a prior given by some experts. Matching of the input text against other texts is performed using Euclidean norm in the term space. Due to the fact that output term base is organized in the term-wise form, text matching is simplified since it is performed only against those texts that the given term belongs to. By using user-friendly interface in the form of if-then rules, the overall system is open for future inclusion of new syntax and semantics rules.