Abstract: 【Purpose/significance】Aiming at the problem that the content-based recommendation method fails to fully consider the complex implicit relationship between papers or between papers and authors, the relationships contained in academic knowledge maps and the content-based method are fully considered, and a cross-domain literature information representation method based on graph comparative learning is proposed by combining these two methods and complementing their respective shortcomings (Graph Contrast Learning for Information Representation Method，GCLIRM). It aims to solve the problems of single method and insufficient information representation in literature.【Method/process】In this method, both heterogeneous graphs and homogeneous graphs are considered. In heterogeneous graph networks, node sequences are obtained using restart random walk algorithm, and node representations are initial⁃ ized using Skip-gram algorithm. A two-level attention mechanism is designed for node importance modeling and meta-path impor⁃ tance modeling to learn node representations of heterogeneous graphs. Secondly, the pre-trained large model ERNIE3.0 is used to model the node features of the same frame, and then the neighbor information is aggregated by GAT and unsupervised training is per⁃ formed by autoencoder. Finally, comparative learning strategies are designed to learn the final node features, that is, document informa⁃ tion representation.【Result/conclusion】The F1 value of GCLIRM method in the two downstream tasks of subject classification and journal classification was increased by 5.31% and 2.49% compared with the sub-optimal solution, respectively, indicating that the method in this paper has greatly improved the representation ability of cross-domain literature information.【Innovation/limitation】 This paper designs a hybrid method to extract literature information representation, which overcomes the problem of inadequate repre⁃ sentation of a single method, and has certain feasibility and accuracy, providing reference ideas for subsequent relevant research.