Research On Personalized Recommendation Algorithms For Photography Image

The recommendation of image is a hot topic in the recommendation field.As a special category in the images,the photographic image has special features in style.In the process of picture feature extraction,the influence of the image style factor needs to be fully considered.On the photographic image sharing platform,users have social relationship attributes which have a great influence on user preferences.In order to make personalized recommendation for users,it is necessary to comprehensively consider image features and user social attributes.Accuracy and diversity are often two contradictory concepts in the evaluation indicators of recommendation results.The research of recommendation algorithms is to achieve a balance point between accuracy and diversity.On the basis of ensuring the accuracy of the existing recommendation list,it is very important to improve the diversity of the recommendation result through the reordering algorithm.In the actual recommendation process,it is necessary to fully consider the difference in user interest,and recommend results with different diversity for users with different interests.Traditional diversity indicators do not accurately measure the diversity of recommendations,and should be comprehensively evaluated in conjunction with user interest.This paper has studied the above problems,and the main achievements are as follows:1.A photographic image recommendation algorithm based on deep learning is proposed and implemented.Based on the latent factor model,the algorithm focuses on the influence of image style,optimizes the target formula of weighted matrix factorization(WMF),and especially increases the proportion of image style in image features,which is a good representation of the photographic image feature.The algorithm learns the image latent factors through the convolutional neural network,and finally calculates the recommended scores of users for images to obtain the recommended list.2.A socialized image recommendation algorithm based on latent factor expansion graph is proposed and implemented.Based on the traditional social network graph,the algorithm establishes a bipartite graph that represents users' preference for images.The algorithm establishes implicit class nodes based on image latent factors.The recommendation process uses the strategy of graph sorting to calculate the ranking of nodes.3.A personalized reordering algorithm for recommendation list oriented to diversity is proposed and implemented.The concept of user interest is proposed in the algorithm,and the reordering strategy is performed for each user's recommendation list according to user interest,so as to improve the diversity of recommendation results.At the same time,a new personalized diversity evaluation indicator is proposed,which makes the diversity of the recommendation list personally match the user interest well.