Research And Application Of E-commerce Hybrid Recommendation Algorithm Based On GRU Network

In recent years,personalized recommendation of e-commerce has become a research hotspot in the field of e-commerce.At present,although personalized ecommerce recommendation technology has achieved good results,many problems such as cold start,diversity of recommendation results,high efficiency of recommendation algorithm still restrict the development of personalized recommendation technology.With the development of recommendation and hybrid recommendation technology based on users' short-term session data in the field of e-commerce personalized recommendation,researchers can use deep learning technology to analyze users' Multidimensional historical data to get products that users may be interested in.This paper focuses on the historical data of users and commodities of e-commerce platform,and discusses the algorithm of e-commerce personalized hybrid recommendation based on deep learning.In this paper,a hybrid recommendation model HGRU based on GRU network and matrix factorization is proposed,and attention mechanism is introduced to obtain candidate recommendation list.Then,HGRU-GAN is constructed to reorder the candidate recommendation list.Finally,the design and implementation of the e-commerce recommendation system which integrates the hybrid recommendation algorithm proposed in this paper are completed.The specific work is as follows:(1)Most of the existing recommendation algorithms rely too much on the user's historical information,and can't achieve good recommendation results in the case of sparse data.This paper presents a hybrid e-commerce recommendation algorithm HGRU.This algorithm improves the original GRU recommendation algorithm,introduces the matrix factorization algorithm to obtain the user's long-term information,and introduces the attention mechanism to judge the user's main purchase intention.The long-term implicit factors of users and goods are obtained by decomposing the userproduct rating matrix.GRU network obtains users' short-term purchase interest by analyzing users' click behavior.On the one hand,HGRU algorithm makes up for the deficiency of GRU recommendation algorithm for users' long-term information acquisition ability,on the other hand,it makes an improvement on the problem that the recommendation algorithm is interfered by users' irrelevant operations.(2)In this paper,we further optimize the HGRU algorithm's recommendation list sorting problem.Although HGRU has made some progress compared with advanced algorithms such as GRU-Rec in the average reciprocal ranking,there is still room for improvement in the ranking of recommendation list because human behavior is affected by many factors.This paper proposes a generative adversary network model,HGRUGAN.This model aim to enhance the learning of HGRU algorithm for product relevance score,so that the recalled products in the recommendation list are more advanced.HGRU-GAN takes HGRU as generator and designs a symmetric network as discriminator.The discriminator tries to distinguish the real high score goods in the recommendation list generated by the generator,and the generator tries to cheat the discriminator to produce a better recommendation list.Compared with HGRU algorithm,the MRR of HGRU-GAN on YOOCHOSE1/3 dataset and DIGINETICA dataset increased by 4.8% and 5.2% respectively.(3)The design and implementation of e-commerce platform is completed,and the hybrid recommendation algorithm proposed in this paper is applied to the platform.Personalized e-commerce recommendation platform uses big data technology to build.Yes,e-commerce platform can deal with massive data storage and preprocessing.