The use of fuzzy search methods when identifying input messages inevitably leads to the appearance of quite a lot of information noise in the output data. It is proposed to solve the current problem of cutting off information noise in the output data by analyzing the results of a fuzzy search using a dictionary of paronyms for nominal components. In relation to nominal components, paronyms are those nominal components that, on the one hand, satisfy the criteria of similarity of an information retrieval system, but, in fact, differ in meaning (they relate to completely different individuals). For the first time, the problem of reducing information noise in the output message of information retrieval systems was solved using an information filter, the structure-forming element of which is a dictionary of paronyms of nominal components. When compiling a dictionary of paronyms, a mechanism was used to categorize paronyms of nominal components depending on the number of distorted characters. The proposed approach to reducing information entropy depending on the category and number of paronyms of nominal components in family-name groups can be applied not only for information search by family-name groups, but also in a broad sense – for text data frequently used in queries, which are subject to various distortions.