As a new computing technology proposed with the development of 5G, IoT technologies and increasing requirement of mobile applications and services, edge computing enables mobile application developers and content providers to serve context-aware mobile services (e.g., mobile app recommendation). Mobile app recommendation is known as an effective solution to overcome the information overload in mobile app markets. Most existing models only consider user-app interaction and feature modeling, and neglect the structural information which actually is a crucial part in the scenario of app recommendation. To fully exploit both structural and feature information for app recommendation, this paper proposes a novel heterogeneous graph neural network framework (HGNRec) including one inner module and one outer module. Specifically, the inner module is able to use a node-level attention to learn the importance between a node and its meta-path based neighbors. The outer module with a path-level attention can learn the importance of different meta-paths. With the learned importance from two modules, the comprehensive embeddings for user and app nodes can be generated by integrating features from meta-path based neighbors. Extensive experiments on the real-world Google Play mobile app dataset demonstrate the effectiveness of HGNRec.

随着5G，物联网技术的发展以及对移动应用程序和服务的日益增长的要求，边缘计算作为一种新的计算技术而提出，边缘计算使移动应用程序开发人员和内容提供商可以为上下文感知的移动服务（例如，移动应用推荐）。推荐移动应用程序是克服移动应用程序市场中信息过载的有效解决方案。大多数现有模型仅考虑用户与应用程序的交互和功能建模，而忽略了结构信息，而结构信息实际上是应用程序推荐场景中的关键部分。为了充分利用结构信息和功能信息来推荐应用程序，本文提出了一种新颖的异构图神经网络框架（HGNRec），该框架包括一个内部模块和一个外部模块。具体而言，内部模块能够使用节点级别的注意来了解节点与其基于元路径的邻居之间的重要性。具有路径级别关注的外部模块可以了解不同元路径的重要性。从两个模块中学到的重要性之后，用户和应用程序节点的全面嵌入可以通过集成基于元路径的邻居的特征来生成。在真实的Google Play移动应用数据集上进行的大量实验证明了HGNRec的有效性。