Objective: Safety culture plays an important role in safety management in the construction industry. Safety culture evaluation items form an important basis for determining the limitations of safety culture and designing measures for improvement. However, the existing safety culture evaluation items are simply expressed and backed by insufficient logic. Moreover, they lack the management mechanism perspective, thus affecting practical guidance. To manage safety culture in the construction industry more effectively and improve construction safety in engineering practice, it is necessary to link safety culture with the management mechanism. Methods: There are two main analytical methods used in this study: management mechanism theory and semantic analysis method. The function of the management mechanism is to make the managed objects move in the direction and interact under certain rules desired by the manager, thus ensuring proper management. Semantic analysis refers to the use of various methods to learn and understand the semantic content of the text. Semantic role labeling (SRL) is a method to analyze the relationship between sentence components and predicates, with sentence predicates as the core. Common semantic roles include agent, sufferer, time, place, etc. The semantic analysis results obtained by SRL are highly consistent with those of the management mechanism. Therefore, the SRL method is used to analyze the management mechanism elements and connotations of the safety culture evaluation items on the semantic analysis level. Based on the management mechanism theory and semantic analysis, this study first collects a total of 1 [KG-*7]107 evaluation items from 33 construction safety culture evaluation-related articles. Further, based on the semantic role of the evaluation items, eight types of management mechanism elements are extracted, including agent, behavior, property, time, location, manner, range, and reason. By summarizing all interactions between the management mechanism elements in the evaluation items, the rules of safety culture management mechanism can be explored. Results: The extraction result analysis indicates that the safety culture evaluation items can be characterized using two management mechanism meta-models and their extended forms, which include one subject or two subjects. Additionally, the management mechanism elements of agent, behavior, property, time, location, manner, range, and reason and their interactions can also be characterized using these models. Of them, all kinds of management mechanism elements not only involve a diversified management practice connotation but can also be adjusted to form the design basis for safety culture improvement measures. The management mechanism elements are also clearer and more practical compared with the conventional brief description of safety culture evaluation items. In addition, this study presents the construction industry safety culture improvement strategy based on the perspective of management mechanism, and the application of this strategy is explained using an actual case. Conclusions: On the basis of the conventional evaluation-oriented safety culture promotion method, this study theoretically enriches the connotation and characteristics of the safety culture management mechanism and provides suggestions for construction industry employees to implement safety culture management in practice.