Currently, consumers are increasing the demand for personalized fashion recommendation systems, especially in e-commerce. The success of fashion recommendation largely relies on the fashion design knowledge base, which is the modelling of the professional knowledge and experience of fashion designers. As fashion products are rather emotional and easily influenced by fashion trends, fashion design knowledge bases always involve rich fashion-style words and various garment component elements, which results in conflict rules in personalized fashion recommendations. For system users, the recommendations cannot fully meet their personalized needs. In this context, this article proposes a novel garment design knowledge base by integrating the conflict rule processing mechanism and its application to the personalized fashion recommendation system. The proposed research methodology of the processing mechanism is based on subjective evaluation and fuzzy logic. Firstly, through the subjective evaluation by an expert group, the relationship between fashion-style words and garment component elements is established. After that, the conflict categories are identified and then the relationship between conflict categories and garment component elements is established. As the experimental data obtained from the subjective evaluation are semantics, fuzzy logic is employed to quantify and model the relationships. Finally, the established processing mechanism of conflict rules is applied to establish a personalized dress recommendation system in order to validate its performance. The experimental results show that the proposed mechanism is feasible and able to enhance the success of the recommendation, which can be further applied to the recommendation of all kinds of garments and other industrial products.