In the fields of data mining and artificial intelligence, rule learning aims to extract general rules from data to facilitate the prediction and classification of new instances. Language models, especially large language models (LLMs) pretrained on extensive text corpora, have demonstrated remarkable capabilities in capturing complex semantic structures and logical relationships inherent in natural language. However, effectively leveraging these models for rule learning involves several critical challenges, such as balancing interpretability and generalization, efficiently integrating symbolic reasoning with neural methods, and systematically evaluating learned rules. This survey provides a comprehensive overview of recent advances in applying neural network-based and large language model-based methods to rule learning. We categorize current approaches, critically analyze their strengths and limitations, and discuss promising strategies and research directions to address existing challenges. The findings of this study offer valuable insights and guidelines for future research in enhancing rule learning capabilities using language models, contributing to the development of more interpretable and robust artificial intelligence (AI) systems.
Cheng et al. (Sun,) studied this question.