Abstract: 【Purpose/significance】The study aims to utilize book text content for recommending similar books and to enhance the effi‑ ciency of book similarity calculations in the context of massive book data environments.【Methods/process】Key phrases were ex‑ tracted from the book's text content, and their TextRank values were calculated within the phrase network. Book vectors were then es‑ tablished and combined with the Hierarchical Navigable Small World (HNSW) algorithm to determine the similarity between target and recommended books.【Results/conclusion】The average accuracy of user evaluation obtained by using the content-based similar book recommendation method reaches 0.807, which is significantly higher than that of TF-IDF and TextRank-based textual represen‑ tations. This method achieves a better book recommendation effect. Moreover, the HNSW algorithm reduces computational efficiency to a logarithmic level, optimizing the calculation of similar books in big data environments.【Innovation/limitation】This study innova‑ tively combines graph structures and HNSW algorithms to improve the accuracy and computational efficiency of book recommenda‑ tions, but its reliance on Tencent's dictionary affects the universality and cross-language adaptability of vector expressions.