The invention relates to a construction method of network information classification system, which integrates the advantages of the existing two classification systems, designs a new construction method, maps words to a low-dimensional vector space by Synonyms, measures the similarity of words by the distance of the word vector space, and compensates the lack of single one-hot word vector semanticinformation so as to obtain more comprehensive similarity categories of words; The traditional literature classification method and self-compiled classification method are used to make up for the shortcomings of the lack of adaptability in the field of literature classification and self-compiled classification method is not scientific so that the classification system of science specialty can beconstructed. Moreover, it proposes to combine user's business requirements and preset screening rules, to interpret the domain terminology association, to select, delete, add, modify and construct thefinal classification system by referring to the prompt, so as to make the classification system more scientific, accurate and professional.