A Novel Deep CNN Method Based on Aesthetic Rule for User Preferential Images Recommendation
Jing Yu, Lulu Zhao · Journal of Applied Science and Engineering · 2021
ABSTRACT How to quickly and effectively search for users’ preferential images among numerous image information resources has become an important problem to be solved in the field of image recommendation. The traditional method improves the performance of personalized image search, but the descriptive information or label setting for the image is subjective to the user. In this paper, we propose an aesthetic rule-oriented deep CNN (convolutional neural network) method for user preferential images recommendation. The image selected by the user is used to represent the user’s preference, and then the deep CNN is adopted to extract the image features to fully capture the inherent diversity of the image. Meanwhile, the image aesthetics rule calculation method is also used to calculate the aesthetic characteristics of the image to ensure that the user’s aesthetic characteristics are satisfied. Experiments show that this proposed algorithm is an effective recommendation method for aesthetic preference.