Remote Sensing Image Recommendation Using Multi-attribute Embedding and Fusion Collaborative Filtering Network.

With the popularization of remote sensing applications, the number of remote sensing users and their application requirements have increased explosively. Introducing recommendation systems into the distribution of remote sensing images can lower the threshold to obtain images for the public and subvert the traditional image service pattern. Different from other industries, users in the remote sensing field do not purchase data frequently in most cases, making it difficult to rely solely on order information to make recommendations. Therefore, how to use the semantic information of users and images as a foundation is an urgent problem. In this paper, we propose a feasible framework for personalized remote sensing image recommendation. We first describe remote sensing users and images through user duties, image semantics and order information; then, we complete the information modeling process via a knowledge graph. Next, we extract multidimensional features from the knowledge graphs of users and images with the help of knowledge representation learning and quantifiable features. Finally, using the high-dimensional spatial modeling capabilities of deep neural networks to perform a deep interaction exploration, we propose a Multi-attribute Fusion-based Collaborative Filtering Network, called MaF-CFNet. The model effectively fuses the multidimensional attribute features of users and images and obtains a recommendation score for each candidate image. The experimental results on our datasets demonstrate the effectiveness of MaF-CFNet compared to traditional methods.