Research On Clothing Personalized Recommendation Based On User Preference And Image Content

With the rapid development of fashion shopping platforms such as Taobao,Mushroom Street and Vipshop,although more and more clothing products are available for users to choose from,the problem of information overload has followed,and users want to select the right clothing.It takes a lot of time and effort.Nowadays,the most widely used personalized recommendation algorithm is the Collaborative Filtering(CF)algorithm.Although it has been able to produce very good recommendation effects,it has insufficient recommendation accuracy,low coverage,and cannot fully exploit the potential interests of users.The problem,secondly,user-based collaborative filtering ignores relevant information about the apparel product itself,such as the visual characteristics and attribute characteristics of the garment,and the compatibility between the apparel products.In view of the above problems,the main research work of this paper is as follows:(1)Aiming at the problem that the recommendation method of clothing is not accurate enough to fully exploit the potential interest of users,an improved personalized recommendation algorithm based on user preference for clothing attributes is proposed.Firstly,a deep convolutional neural network is constructed to realize clothing attribute extraction for clothing image content.On this basis,user attribute vector is constructed according to the extracted clothing attributes,and the similarity based on user attribute preference is combined with user interest preference similarity based on time factor to construct the final user preference model.Compared with the user-based collaborative filtering algorithm UCF,and the project-based collaborative filtering algorithm ICF and the literature-based collaborative filtering UCSVD algorithm based on project preference,the accuracy of the UIACF algorithm can be increased by up to 14%(2)Most of the existing collaborative filtering recommendation algorithms consider fine-grained modeling from the user’s point of view,and ignore the influence of matching compatibility between clothing products on user purchasing decisions.A clothing joint personalized recommendation algorithm based on BPR fusion user and clothing collocation is proposed.The unsupervised convolution auto-encoder is used to apply the clothing image content feature to the clothing product matching compatibility relation matrix,and the time factor is added to the user item matrix to establish a user-clothing-time based relationship matrix.The user-clothing-time interaction relationship and the collocation relationship between the clothing products are simultaneously used as the auxiliary information for recommendation,and the user-clothing-time-matching relationship matrix is constructed.Finally,the Bayesian personalized sorting algorithm is used to realize the fusion modeling.It shows that the UTCBPR can effectively improve the performance of recommendation.(3)Personalized clothing customization and recommendation system.In order to more effectively and accurately meet the personalized needs of customers,a personalized clothing customization and recommendation system has been established.The recommendation algorithm proposed in this article is applied in the system.The application results show that the service is provided in the city,province,and country,and the overall sales The trend has improved,with a cumulative increase of 21.16%in new users,of which consumers and designers have increased by about 30,000.