Research On Personalized Clothing Matching Recommendation Method Based On Knowledge Graph

As the younger generation of consumers gradually becomes the mainstream consumer group,personalized demands and a "me-centered" consumption concept have emerged,making personalized consumption a new trend.The prosperity of e-commerce culture has led to an explosive growth of clothing data,making it difficult for consumers who lack professional matching knowledge to efficiently select suitable clothing.Therefore,personalized clothing recommendation has emerged as a recommendation method based on users’ personalized aesthetics and needs,which can meet consumers’ constantly changing fashion demands while improving the conversion rate of product sales,thereby bringing better commercial benefits to clothing shopping platforms.However,current personalized clothing recommendation algorithms often ignore the incomplete clothing attribute information,and fail to integrate user preference features and clothing attribute information well.As a kind of semantic knowledge base,knowledge graph has the ability to represent and organize data graphically.In the field of apparel recommendation,constructing a graph of apparel information in the form of a triad can clearly describe the relationship between apparel and attributes,and through the analysis of the knowledge graph,the relationship between user requirements and characteristics of apparel can be better understood,and the entities and relationships in the knowledge graph can be intuitively represented as a graphical structure,which facilitates the visualization and interpretation of recommendation results.Therefore,this study proposes a knowledge graph-based personalized clothing recommendation method and designs and develops a corresponding personalized clothing recommendation prototype system.The main work of this thesis is as follows:(1)Construction of clothing knowledge graph.To construct the clothing knowledge graph,relevant books,literature,and expert knowledge in the field of clothing are summarized and analyzed to determine clothing attribute elements and construct a model layer for the clothing knowledge graph.Clothing data is obtained from shopping websites for knowledge extraction.The data is enhanced and expanded,and the knowledge is fused to obtain a triplet representation of clothing attribute knowledge.The knowledge is then stored to populate the data layer and ultimately build a complete clothing knowledge graph.(2)Personalized clothing recommendation algorithm based on knowledge graph.On the user-clothing interaction graph,a Light GCN graph convolution neural network lightweight model is employed to decompose the interaction matrix between users and clothing,learn the similarity information between users and clothing,and update the vector representations of users and clothing.Then,a self-attention mechanism is used to extract users’ clothing preference features,which are further used to implement personalized clothing recommendation.The obtained user personalized preference features are used to optimize a hybrid recommendation algorithm based on the knowledge graph for scenarios where clothing attributes are missing in clothing recommendation.This enhances the personalization and accuracy of clothing recommendation,while improving the interpretability of the algorithm.(3)Implementation of personalized clothing recommendation system.Based on the personalized clothing recommendation algorithm proposed in this study and combined with front-end and back-end development technologies,a personalized clothing recommendation system has been designed and implemented.The system enables clothing merchants to maintain clothing knowledge graph data when they put their products online and can provide personalized clothing recommendations to consumers quickly.