Research On Personalized Clothing Recommendation Model And Algorithm Based On Deep Learning

At present,e-commerce has developed into a mainstream way of shopping,and the emergence of recommendation system has alleviated the problem of information overload to a certain extent.As an important commodity category of online shopping,how to achieve efficient clothing matching recommendation has become a research hot spot in industry and academia.In this thesis,based on the analysis of user's personalized preferences,feature extraction of deep learning,personalized clothing matching recommendation technology and algorithm,the personalized clothing matching recommendation algorithm and method based on user's history are proposed.The main work of this thesis includes:(1)A recommendation algorithm for personalized clothing matching has been proposed.CNN has been used to extract and label the user's personalized features,and the feature tags have been counted and sorted to highlight the influence of user preference features on the recommendation results,so as to realize personalized recommendation.This thesis used the clustering algorithm to cluster the products in the candidate set according to the characteristics,and constructed the clothing combination graph and feature combination graph according to the user's historical purchase records and clustering results,so as to score and recommend the products in the candidate set.Compared with the existing clothing matching recommendation algorithm,the accuracy and recall rate have been improved.(2)A method to expand the user's historical purchase record based on DCGAN has been proposed.DCGAN has been used to train user's historical purchase records and generate images similar to users' historical purchase records,so as to achieve the purpose of data augmentation and feature filtering.The generated image and the user's historical purchase records have been used as the basis of personalized feature extraction to maximize the value of personal goals.The quality of the generated image and the similarity of the user's historical purchase record image are verified by the evaluation index,which proves that the method can be used to expand the user's historical purchase record.(3)A method of extracting color feature labels based on deep learning network has been proposed.Through the research of image color feature and shape feature extraction method,after using deep learning technology to train the feature extraction model,the trained model has been used to extract color and shape feature labels from the candidate set and user history purchase records,with the help of VGG16 neural network model,it has been simplified and improved,and a lightweight neural network C-VGG suitable for color feature extraction has been obtained.Compared with VGG16,this method improves the accuracy and training efficiency.The recommendation algorithm proposed in this thesis can be widely used in clothing sales websites and virtual fitting equipment to improve the shopping efficiency of users.