Clothing Style Recommendation System Based On Hybrid Model

With the rapid development of e-commerce platforms,online shopping has become an indispensable part of people's daily life.With the increase in the variety and quantity of clothing,people need to spend more time looking for their favorite clothing products,so a good clothing recommendation system becomes more and more important.Traditional recommendation algorithms generally have problems such as cold start,sparse data,and poor stability,and mainly rely on the user-product evaluation matrix,which cannot process image data.Based on the existing recommendation system,this paper conducts the following researches on the above problems:1)A style preference dataset is constructed.Based on the pictures and POG data sets of ecommerce platforms such as Taobao and Jingdong,the classification and selection are carried out,and three clothing style definition standards are adopted.Firstly,according to the characteristics of the times and national traditions,get hip-hop style and ethnic style;secondly,according to the characteristics of fabrics,colors,patterns,etc.,get classic style and Chanel style;finally,according to the functional characteristics of clothing,get sports style and workplace style.A clothing style dataset of 7200 images with six distinctly representative styles is built.2)Clothing style recognition algorithm.This paper combines convolutional neural network and Transformer to build a hybrid model to deal with the task of clothing style recognition.Specifically,the improved Res Net-50 convolutional neural network is used to extract image features,the learning and generalization ability of the model is improved by using the inductive bias of the convolutional neural network,and then the attention mechanism is introduced to learn the correlation between global features degree,and use transfer learning to improve the performance of the model on small datasets.Experimental results show that our hybrid structure improves the Top-1 accuracy of clothing style recognition by 3.2% compared to the traditional Res Net-50.3)Clothing style recommendation algorithm and system implementation.In this paper,a recommendation algorithm based on k-means is proposed,which extracts the features of the pictures uploaded by the user to obtain the clothing style feature vector,then clusters the feature vectors of the same style category,and finally calculates the cosine between the user input picture and the images in the same cluster.Similarity,you can get the clothing image that is most similar to the style of the user input image.The article designs and implements a clothing style recommendation system.This system is oriented to all users who log in to the platform.The system consists of two modules: clothing style recognition module and clothing style recommendation module,which provide the functions of uploading images,clothing style recognition,and clothing style recommendation.Users only need to upload an image of clothing to get the style category of clothing,and the system will automatically recommend other clothing images of the same style to the user.To sum up,for the clothing style recommendation problem,this paper builds a style preference data set,designs and implements a clothing style recommendation system.The system uses clothing style recognition algorithm to extract clothing image features,and then uses clothing style recommendation algorithm to give recommendation results.It can meet the needs of users to identify clothing styles and recommend clothing.Experiments show that the system has the advantages of high recognition accuracy and fast recommendation speed.