Research On Application Of Recommendation Technology Based On Knowledge Graph Of Clothing Field

With the rapid development of the clothing industry and the diversification of clothing brands,how to make clothing accurately and intuitively presented in front of the user is crucial.The recommendation system allows users to spend less time to find the clothes they are interested in and suitable for them.In order to improve the accuracy of the recommendation system,the introduction of some auxiliary information into the recommendation algorithm can effectively alleviate the problems of cold start and sparsity,and the knowledge graph can also be used as a kind of auxiliary information into the recommendation system.Therefore,in-depth analysis and research were conducted on how to integrate apparel knowledge graph into the recommendation algorithm as auxiliary information,including:(1)The entity recognition algorithm of CNN-BNLSTM-CRF is improved to identify named entities in the apparel field,which improves the confidence of apparel knowledge graph and solves the limitation of entity extraction.First,the effective features of word vectors are obtained by using the convolutional neural network,then the context features of words are learned by using the bidirectional long and short term memory network based on nonlinear transformation,and the vector representation of the context of each word is output.Finally,the optimal tag sequence of sentences is obtained by using the conditional random field model through self-learning.The experimental results show that the improved fusion model is better than the single and double models in recognition,achieving 88.95% accuracy,93.56% recall rate and 91.20% F1 value.(2)In order to improve the accuracy of clothing recommendation and alleviate the data sparseness and cold start problems of traditional clothing recommendation algorithms,this paper combines the clothing knowledge graph with the collaborative filtering recommendation algorithm,and proposes the CTransR-CF algorithm.Firstly,the obtained apparel entities and their relations are vectorized by the improved CTransR knowledge representation algorithm,and the semantic similarity matrix of apparel entities is obtained.And then it is fused with the similarity matrix based on user behavior,establish regression model and finally it is sorted according to the similarity degree to get the recommended list of clothes.Finally,through comparative analysis of experiments,the CTransR-CF algorithm proposed in this paper improves the accuracy of the algorithm according to different nearest neighbor Numbers selected,which is better than other recommended algorithms.(3)At last,the algorithm proposed in this paper was further verified,and a recommendation system based on knowledge graph of the apparel field was realized by designing for the demand analysis and architecture of the system.The system is mainly divided into clothing knowledge graph building blocks and clothing recommendation module,which in the process of knowledge graph construction of clothing,designed the semantic type of clothing domain knowledge graph and its relationship,at the same time to the clothing domain ontology library on the construction of knowledge graph,clothing entity recognition implementation and the effect of the storage of knowledge graph of clothing,clothing knowledge query and visualization,etc.The clothing recommendation module trains the proposed CTransR-CF algorithm model,makes score prediction based on user information,and recommends according to the degree of similarity.This paper designs and implements a recommendation system based on apparel knowledge graph.It uses top-n collaborative filtering recommendation algorithm based on feature learning of knowledge graph to improve an entity recognition algorithm oriented to apparel knowledge graph to obtain entity semantic features in the knowledge map,and introduces the knowledge graph into the collaborative filtering algorithm.The system can better meet the needs of users.