Design And Implementation Of Hybrid Recommendation Algorithm Considering User Life Cycle

In the Internet era of rapid development,our consumption on the mobile device is a normal scenario,electricity companies now for the use of marketing tool for the guest,to attract users to enter the platform is not a difficult task,but due to the electric business platform competition is intense,more and more users choose,how to in brief,the fragmentation of the time,stimulate interest in goods within the platform area allows users to quickly.Now more and more enterprises introduce personalized recommendation module in their own e-commerce platform to improve the user’s activity.This paper takes the personalized recommendation algorithm in the recommendation module of e-commerce platform as the research topic.Therefore,this paper studies the development process of e-commerce recommendation algorithm and the recommendation model based on deep learning commonly used nowadays,such as RNN,LSTM and GRU.As user-centered Internet marketing thinking is becoming more and more important in e-commerce ecology,enterprise management planning will be carried out from the perspective of users in the process of personalized marketing and user operation.With the development of personalized recommendation algorithms in the past two years,there are very few researches on the user life cycle concept integrated into business content.Therefore,this paper takes the user life cycle as the perspective to optimize the existing personalized recommendation model of deep learning.Firstly,this paper studies the theories related to the life cycle of e-commerce users,focusing on the classic AARRR model in the life cycle of users.But because AARRR theory for the purpose of this article studies the content of some limitations,so this article is based on the exploration of user behaviors,improved the original life cycle theory,put forward the GMS（Growth Period,Growth Period,the said Period,Mature Period,Sleeping Period-dormant）life cycle theory,make it into electrical contractor recommendation system,better help business enterprises to fit in different life cycle stages of user requirements,improve the accuracy of recommendation.Under the different life cycle stages,the user’s behavior rule,user’s demand to the main tendency is not the same,in G-growth,for example,user behavior with scarcity,need platform to do take the initiative to guide a hot commodity,in the M-mature period,the user’s interest in the goods can be gained through a lot of historical data and in S-sleeping period,then you need to buy some popular goods,add some historical behavior to awaken the user’s interest in platform products.Based on the multi-stage,we study the users to try in different life cycle stage to choose the most suitable matching algorithm model,design a model of hybrid recommendation algorithm based on deep learning framework,user different stages is adapted to different stage model,thus achieve overall recommendation model effect is the best state.GRU is selected as one of the main algorithms in this paper because GRU optimizes gradient disappearance and other problems on the basis of traditional RNN and improves the operational efficiency on the basis of LSTM.In addition,in order to better optimize the problem of reasonable distribution of interest degree in electronic shopping malls,attention mechanism is used in this paper to study the mixed deep learning recommendation algorithm with attention factor.Time information is a key feature of the personalized recommendation system,as is often the case,his paper puts forward a kind of blend in time related information GRU helped attention mechanism model,and through to the user to browse the length of the information goods analysis,cleaning and standardized treatment length factor,the attention and time attention factors and other user behavior data as the input information of deep learning model for training and prediction,more fully recommend effect using the time information to ascend.Finally,in order to verify the effect of the hybrid recommendation algorithm based on the user life cycle proposed in this paper,some single models and hybrid models are adopted for comparison,such as RNN,LSTM,CNN,popular recommendation algorithm and different models used in different stages.the hybrid recommendation algorithm proposed by Dropout is added for further optimization.This paper applies the user behavior data set of the real e-commerce platform APP,which has rich user scale and user behavior data,including more than 3.8 million user behavior detailed data.The training and testing models in this paper are all applied in the recommendation module of this e-commerce app,so as to try to improve users’ interest in the recommended products in this module...