Intelligent Costume Recommendation System Based On Support Vector Machines And Blackboard Models

With the rapid development of the internet technology,more and more consumers enjoy the online shopping,especially in the clothing industry.Therefore,how to give customers a better and more professional online shopping experience has become a research hotspot.The current costume recommendation systems based on the physical characteristics that acquired from the customer ware not accurate.Apparently because the customers can't be sure to position their appearance types.In addition,the current popular costume recommendation recommended goods to the costumers according to their interests and purchase behavior.These systems didn't provide the costumers with the suitable costume according to their actual appearance and lacked the suggestions from the professional person.With the rapid development of deep learning,combining image recognition technology and online e-commerce will become the trend development.Firstly,the costume expert recommendation system(CERS)designed in this paper obtains the customer's photos through human-computer interaction,and then uses the multiclassifier model of Support Vector Machine based on convolution neural network(CNN-SVM)to extract the physical features.Secondly,the customer's appearance information is stored in the fact database and the knowledge source is stored into the rule base in a form of production rule.A dynamic search mechanism is added to the blackboard model to improve the search speed.The costume recommendations of the system are intelligent and personalized based on the different physical characteristics and the experts' experience.Support Vector Machine(SVM)is chosen as the classifier because of SVM has high classification accuracy and generalization ability.This paper takes the face classification as an example and analyses the SDM algorithm and CNN algorithm.The experimental results show that compared with SDM,CNN can directly implicitly extract features without image preprocessing and not lose face profile information.So CNN-SVM multi-classifier is taken as the core technology of customer information acquisition module.The module automatically obtains the customer's skin color,face,shoulder,body characteristics and solve the problems of subjectivity and uncertainty.The costume recommendation system based on the expert system to simulate the thinking process of experts.The recommendation list is obtained by forward reasoning according to the customer's skin color,face,shoulder.The knowledge base uses the production rules to save the customers,clothing information and knowledge.The reasoning machine adopts the blackboard model with dynamic search mechanism to readjust the priority order of knowledge source(KS).This mechanism can ensure the next level KS to be searched in a smaller amount.This method solves the shortcomings of the classic blackboard model in the expert system and effectively improve rule matching and search speed.