Abstract: To control the construction quality of concrete dams, it is necessary to accurately find normative clauses to obtain normative knowledge. Manual finding of these clauses based on the managers’ experiences is inefficient and error-prone. This paper adopts Natural Language Processing technology, from the perspective of sentence semantics, to overcome the shortcoming of considering keyword queries only, and develops a deep retrieval method of construction norms driven by semantic inference. This method combines Bi-GRU and the Transformer structure to improve a MLP-based twin-tower model, and couples it with a BM25 model to realize the intelligent retrieval of the norms; through semantic analysis, it evaluates whether or not the quality inspection project is aligned with the norms. Application to an actual project shows that the accuracy of the first three normative clauses recalled by our coupled method reaches 82.3%, higher than that of the common retrieval method, verifying its engineering applicability. In addition, we have developed a tool to provide an operation platform for promoting intelligent checking of the quality compliance of concrete dam construction.

Key words: norm knowledge retrieval, Natural Language Processing, concrete dam construction, semantic matching, quality control