A large number of real-world graphs or networks are inherently heterogeneous, involving multiple types of nodes and relationships. Heterogeneous network representation learning is to embed the rich structure and semantic information of heterogeneous networks into low-dimensional node representations. The purpose of heterogeneous network representation learning is to find a meaningful vector representation for each node to facilitate downstream task applications, such as link prediction, personalized recommendation, and node classification. With the development of machine learning and deep learning, heterogeneous network representation learning has been widely used due to its powerful modeling capabilities and extensive applications. This paper mainly summarizes the methods and applications of heterogeneous network representation learning. First, the current heterogeneous network representation learning methods at home and abroad are categorized into different groups, including random walk-based methods，second-order proximity-based methods，matrix factorization-based methods，graph attention network-based methods and graph convolutional network-based methods. Then, various models are described in detail by category, and various aspects of the model are explained. The characteristics are comprehensively compared. Secondly, the related applications of network representation learning are introduced, including the recommended system field, the biological field, etc., and the commonly used datasets, open source implementation of heterogeneous network representation learning models are organized for researchers. Finally, the development trend of heterogeneous network representation learning is summarized and prospected. This research provides a comprehensive review reference, as well as some representation learning models and commonly used datasets for scholars in the field of heterogeneous network representation learning.