<p>This study aims to develop an effective recommendation system to assist enterprises in better utilizing patent information during the innovation process. Firstly, key information such as patent titles, abstracts, and application dates is extracted from patent data, and a patent knowledge map is constructed based on this information. This knowledge map uses patents as nodes and forms a complex network structure through citation relationships and domain similarities, providing a data foundation for the recommendation algorithm. Secondly, the RippleNet algorithm is employed for personalized recommendations. This algorithm iteratively propagates ripples to learn the interaction patterns between users and patents, thereby capturing users’ interests and preferences. The goal of applying this recommendation algorithm is to enhance users’ efficiency in obtaining relevant patent information and to support enterprises in making better use of knowledge resources during innovation. The empirical analysis of patent data from listed companies in China further verifies the effectiveness and practicality of the recommendation system. In the knowledge map, after training with the Bidirectional Encoder Representations from Transformers (BERT) model, the recognition accuracy for composition relationships and function achievement relationships in patent abstracts reaches 0.973 and 0.907, respectively, with the accuracy for relative position relationships reaching 0.861. In the personalized recommendation system, as the number of top recommendations increases, the accuracy rate gradually decreases while the recall rate increases. This study explores a personalized patent recommendation method based on a knowledge map, which effectively integrates, stores, and manages a large amount of patent information. It provides enterprises with a more intuitive and convenient knowledge management tool, thereby enhancing their knowledge management capabilities.</p>