In order to better grasp the needs of library users and provide them with more accurate knowledge services, combining the characteristics of university libraries, this article applies library small data to personalized recommendation and proposes a small data fusion algorithm model for library personalized recommendation. This model combines the characteristics of small data and realizes multi-dimensional small data fusion by using fully connected neural network to capture the potential collaborative filtering information between users and projects, better grasp the needs of readers and users, and provide valuable assistance for subsequent personalized recommendation research. The effectiveness of the proposed method in personalized recommendation of library resources is verified by comparing several groups of experiments on public and self-built data sets.