Heterogeneous information network (HIN) has been introduced to characterize complicated and heterogeneous auxiliary data in recommendation systems, called HIN-based recommendations. Developing efficient HIN-based recommendation techniques for both extracting and exploiting data from HINs is challenging. Most HIN-based recommendation methods are based on path-based similarities that cannot completely mine user and item latent structural characteristics. In this project, we are proposing a novel heterogeneous network embedding HIN-based recommendation strategy, called HIN. To embed HINs, we design a random walk approach based on a meta-path to produce significant network embedding node sequences. The learned embedding of the node is first converted by a set of fusion features and then incorporated into an expanded model of matrix factorization (MF). In conjunction with fusion features, the expanded MF model is collectively optimized for the task of rating prediction. The efficacy of the HER model is demonstrated by extensive studies on three real-world datasets. In addition, we demonstrate the HER model’s capacity for the cold-start issue and reveal that the transformed HIN embedding data can enhance the efficiency of recommendations.