The paper relates to the subject field of image recommender systems. The proposed approach of automatic image recommendation addresses the following shortcomings of existing solutions: manual input of metadata by users, the lack of user rating history consideration, significant computational resources. The main idea of the proposed approach is to recognize object classes from images using a convolutional neural network to make recommendations. In the proposed solution users and images are located in the semantic space represented by a graph. Software implementation of the proposed approach and obtained results are considered.
展开▼