Clothing Recommendation System Based On User Style Preference

User style preference refers to the tendency of users to purchase a style of clothing due to their identity,character,occupation and aesthetics while online shopping.The clothing recommendation system based on user style preference aims to recommend clothing that conforms to user style preference to users.Many high-end and luxury clothing brands have formed unique and fixed brand clothing styles in the production and sales process in order to fix consumer groups.Therefore,when consumers choose these brands of clothing,they will not be troubled by the diversity of styles.However,large-scale cheap clothing brands often adopt a multi-style mixed sales strategy in order to expand the consumer group.Such a sales strategy makes it impossible for users to quickly and accurately find their style preferences clothing.Therefore,how large-scale cheap multi-style clothing brands can select and recommend user style preference clothing from its many clothing styles has become an urgent problem to be solved to increase sales.In order to solve the above problems,this article attempts to build a clothing recommendation system based on user style preference,which is divided into two modules: style preference recognition module and style preference recommendation module.The recognition module recognizes user input image style and filters dataset images.The recommendation module is used to perform the Locality Sensitive Hashing coding and Cosine Similarity Calculation on the user input image and the dataset images.The two modules cooperate with each other to recommend clothing that conforms to the user style preferences.The recognition module uses Efficient Net-b3 convolutional neural network as the feature extraction model.In order to realize the training of the recognition module,through literature review and market research,the common user style preferences are identified as five types: classical,neutral,sports,hip-hop and celebrity.Through the analysis of the main factors affecting clothing styles: styles(collar,sleeve and profile),colors and fabrics,these five common user style preferences were quantified and expressed,and based on this,the 5100 clothing images collected by shopping websites such as JD.com and Taobao.com,were screened and classified,and a style preference dataset was created for the training of the recognition module.The recommendation module uses the pre-trained Resnet-50 convolutional neural network as the feature extraction model to extract the features of the user input image and the dataset images,and obtain the feature vectors representing the user input and the dataset images.The dataset of the recommendation module is divided into two types:(1)Filtered dataset images.The images are filtered by the recognition module to filter out clothing that is inconsistent with the user’s input image styles.(2)Unfiltered dataset images.The images are all images of the dataset that have not been filtered by the recognition module,and serve as the control group for this study.In order to speed up the recommendation speed of the recommendation module,a random projection method in the Locality Sensitive Hashing technology is used to encode the feature vectors to obtain the hash value of the corresponding feature vectors.By calculating the Cosine Similarity between the feature vectors that are the same as the hash value of the input image and the feature vector of the input image,clothing similar to the user input is obtained and recommended to the user.This paper verifies that the accuracy of the recognition module is 92.88%.In user research,it was found that the popularity of clothing recommendation system based on user style preferences was 33% higher than that of the control group.This shows that the clothing recommendation system based on user style preference proposed in this paper has better applicability.