Research On Image Recommendation Algorithm Based On Multimodal Heterogeneous Hypergraph

Over the past decades,with the prevalence of intelligent mobile devices and advanced mobile multimedia services,it has become quite convenient to upload photos to these websites no matter when and where.Recently,most media sharing websites collect not only photos or videos but also the metadata of them like location and time of taking photos.With the rapid increase of image information,image search has recently received a lot of attention and became a very active domain,it still remains a challenging problem.One of them is how to use multimodal of different features like image tags,image content feature and image location to do image recommendation.How to use multimodal features to do image recommendation in order to get better result than using single feature is the key point of our essay.We analyze researches of current popular image recommendation algorithms and problems of these algorithms,then we propose an image recommendation algorithm framework based on multimodal heterogeneous hypergraph,we use linear regression of machine learning to do weight learning of hypergraph.Then analyze the result in aspect of complexity and satisfaction in our experiments.Details are as follows:1.Image acquisition and construct multimodal heterogeneous hypergraphWe use dataset collected from social network website.Due to that a single feature cannot give a satisfactory recommendation,we extract diverse feature of images like texture feature,content feature and location,then construct the hypergraph on three dimensions then they are integrated together to involve discriminative information from heterogeneous sources.2.Application of machine learning in image recommendationBased on the multimodal heterogeneous hypergraph we construct after we extract features from images,we use linear regression of machine learning to train the weight of every hyperedge.When user gives an input image,we can give a rank for images in our dataset then return the k most similar images to user.3.Result analysis of image recommendation algorithm frameworkWe implement the framework we proposed,then give the complexity analysis and comparison with main popular image recommendation,multimodal image recommendation on the aspect of running time and user's satisfaction.The proposed system architecture consists of two key components:1)offline multimodal hypergraph construction and 2)online search.Via the analysis and comparison of our experiment result,the proposed image recommendation algorithm based on multimodal heterogeneous hypergraph can give a high satisfaction result.