Recommender systems are ubiquitous and play an important role in most service-based applications. Since rich semantics lie in interactions between user and item, various models based on Graph Neural Network (GNN) are widely adopted to model the interactions and achieve excellent recommendation performance. Nevertheless, these models mainly learn nodes in the path between user and item separately. The path is also selected randomly, which would bring in low-quality paths and lead to suboptimal performance. To address the above issues, this paper proposes a model named Multi-View Attention neural network with high-quality paths (HP-MA). The path between user and item is viewed as an entirety, and a neural network is designed to distill path semantic embedding. Meanwhile, a multi-view attention model is provided to learn the influences between embedding of the path and user/item. Besides, we provide a method to construct HP-MA between target user and item, which would consider the length and inner details of the path. Moreover, a generalized version of our model (HP-MA*) is proposed to consider several kinds of multi-hop paths, simultaneously. Lastly, extensive experiments are conducted on several public datasets, the results demonstrate that our model can achieve better performance than the state-of-the-art models.