Using Deep Structured Semantic Model to Analysis Text Documents in the Building Normative Base

This paper analyzes artificial neural networks that can been used to search for web documents in the electronic database of regulatory documents in the field of construction and building materials. The expediency of using artificial neural networks of the category Deeply Structured Semantic Models to solve the problem of semantic analysis of text documents contained in the regulatory framework of buildings has substantiated. Deeply structured semantic models perform a nonlinear projection to map a query into a common semantic space. After such a mapping, the relevance of each document found on the query has calculated by the cosine of the angles between the vector query model and the vector document model. In addition, the architecture of a deeply structured semantic model uses hidden layers that has designed to resize input vectors. This allows models to manipulate vectors of different sizes. The scheme for identifying different documents on the same issue has proposed. The possibility of applying models and methods of fuzzy mathematics to formalization of texts of building norms and rules and expression of their semantics in the internal language of the Semantic Text Information Analysis System has shown.