Knowledge-enhanced AI models for domain-specific application: a survey

With the rapid development of artificial intelligence (AI), deep learning models have played an irreplaceable role across various domains. Through their ability to process and analyze massive datasets, these models enable automated information extraction, pattern recognition, and decision support, thereby enhancing system efficiency and accuracy. However, in practical applications, these models often suffer from limitations such as poor transferability, limited explainability, susceptibility to overfitting, and high update costs. These limitations restrict their large-scale deployment in critical domains and hinder their stability and generalizability in complex scenarios. Recently, knowledge-enhanced models have emerged as a promising solution. By providing auxiliary information that is closely aligned with real-world facts, explicitly defined, and task-relevant, knowledge can effectively overcome these limitations and facilitate the broader application of deep learning models across diverse areas. To the best of our knowledge, this survey focuses on explicit knowledge that is easy to represent and retrieve, providing a clear classification and definition of explicit knowledge for the first time. Based on this foundation, we conduct a systematic synthesis review of the techniques employed in explicit knowledge-enhanced models. This review provides a detailed overview of the mainstream technologies used in various types of explicit knowledge-enhanced models and discusses their domain-specific applications. This review proposes a taxonomy based on three phases of model execution (pre-execution, in-execution, and post-execution) and the characteristics of different types of explicit knowledge. Finally, this review discusses four promising future directions for explicit knowledge-enhanced models.