In order to effectively improve the accuracy of the recommendation algorithm, a modified Latent Dirichlet Allocation (LDA) user interest model is proposed for personalized book recommendation. Firstly, on the basis of the borrowerborrower score matrix, the method of book content similarity calculation is improved by adding the similarity calculation of borrowers’ feature information and similarity calculation of borrower-book attributes. Then LDA theme mining model is used to realize personalized book recommendation, and the corresponding parameter estimation process is given. The experimental results show that compared with the traditional algorithm, the proposed algorithm has high accuracy, which is effective in book mining and can recommend individualized books in which the borrowers have potential interest.