Graph Neural Network and Multi-view Learning Based Mobile Application Recommendation in Heterogeneous Graphs

With the popularity of smartphones, mobile applications (mobile apps) have become a necessity in people's lives and work. Massive apps provide users with a variety of choices, but also bring about the information overload problem. In reality, the number of apps that users have used is very limited, resulting in a very sparse interaction matrix between users and apps. It is not accurate enough to use a sparse interaction matrix to predict numerous unknown ratings, so that the recommended results cannot satisfy users. This paper aims to exploit the user's historical behavior data and the app's side information to make app recommendation to solve the problem of information overload. Specifically, first of all, multiple semantic meta-graphs are designed by leveraging the user information, app information, user historical usage record information, and app's side information. Then, similarity matrices between users and apps based on different semantic meta-graphs are obtained. The graph neural network with the attention mechanism is employed to learn the collaborative information between users and apps, and to selectively aggregate the feature information of the neighbors. Finally, the multi-view learning and attention mechanism are adopted to obtain users' ratings for apps from different perspectives. Comprehensive experiments with different numbers of training samples show that the proposed method outperforms other comparison methods.