Research On Personalized Recommendation Of Fruit And Vegetable Commodities Based On Meta Learning

With the rapid development of e-commerce,more and more people choose to buy fruits and vegetables online.The explosive growth of online information of fruit and vegetable commodities has brought the problem of information overload and affected the user experience of consumers.As a tool to alleviate "information overload",recommendation system has been widely used in recent years.Due to the limited interactive data between users and items,the traditional collaborative filtering recommendation methods have some problems,such as sparse data and cold start.Compared with clothing,books and other commodities,in the actual recommendation scenario,users interact less with fruits and vegetables,and the problems of data sparsity and cold start are more serious.How to improve the recommendation effect of fruit and vegetable commodities through limited interactive data is a subject worthy of research.As a new learning paradigm,meta learning can quickly learn new knowledge and skills through a small number of training samples.It has certain advantages in alleviating the data sparsity and cold start problems in the recommendation process.In recent years,it has been more and more applied to the research of recommendation system.The research on personalized recommendation of fruit and vegetable commodities based on meta learning in this dissertation is to apply meta learning technology to the recommendation of fruit and vegetable commodities,so as to alleviate the problems of sparsity and cold start of sample data.The main research work of this dissertation is as follows:(1)This dissertation proposes the item and user embedded enhancement module.Meta network is a model structure based on model meta learning method.It has the ability to quickly extract some relationship in the task and realize the cross task transfer of knowledge.It can obtain the initial embedded representation of goods and the embedded representation updated by rich interactive information.There is a correlation between them.This dissertation designs a meta network model structure to realize the rapid transformation of embedded feature space of fruit and vegetable goods;Using the user’s comment text data on fruit and vegetable commodities,the semantic features of users are extracted,and the feature embedded representation based on the real preference of each user is generated.Experiments show that the meta network is effective in fruit and vegetable commodity feature extraction and user feature extraction model.(2)This dissertation proposes a learning paradigm Meta EE for meta learning of fruit and vegetable commodities.Meta EE uses its embedded enhancement module of fruit and vegetable commodities and users,as well as the model independent meta learning algorithm to update the parameters of the recommendation model,standardize the overall process of fruit and vegetable recommendation,and guide the parameters of the recommendation model to be updated quickly and iteratively according to the characteristic information of fruit and vegetable commodities and users.(3)Relevant experiments on real data sets verify the effectiveness of the Meta EE learning paradigm proposed in this dissertation.The experimental results show that in the aspect of fruit and vegetable commodity recommendation,the Meta EE meta learning method proposed in this dissertation is used to train the parameters of different recommendation models.Compared with the traditional recommendation technology,its recommendation effect has certain advantages,which is helpful to alleviate the problems of data sparsity and cold start in the recommendation process.