Research On Recommendation Method Based On Dynamic Graph Model In E-commerce

With the popularization of computer networks and the improvement of people's living standards,more and more users access the Internet through various devices,such as mobile phones,Ipads,and smart wearable devices.Enterprises can improve user friendliness by collecting and analyzing user data.In the e-commerce industry,due to the diversity of items,there will be a wide range of items information on the platform.Under the premise that users do not have clear requirements,users cannot find items that they need to purchase in the face of massive amounts of items.The e-commerce platform cannot push items to the target users who are interested in the items.How to better allow users to buy their favorite items and how to better serve users has become the main problem now.Regarding the issue,the recommendation system came into being.The recommendation system is the tool between the user and the item.In the face of massive data information,the recommendation system is an efficient tool for solving data overload.It can reasonably recommend the user's favorite items or information to the user when facing the massive data.Enterprises analyze the user's historical data to discover the user's preferences and the user's potential spending power.In combination with the characteristics of the item or information,the item is recommended to the target user who is interested in the item.Such a recommendation system can not only help users find items they like,improve user-friendliness,but also help the company provide better services,and it also helps companies recommend items to target users.We recommend based on the user's preferences,and the user's preference changes over time.The user's long-term and short-term preferences are very important to the user.The use of time factors can be a good description of long-term and short-term preferences.Therefore,this dissertation proposes a user preference calculation method based on dynamic graph model,and proposes two kinds of recommendation algorithms based on user preferences.The main contents of this paper are summarized as follows:(1)First,a user preference calculation method based on dynamic graph model under the MapReduce framework is proposed.This method obtains the user's final preference through two rounds of MapReduce calculations.In the first round of processing,the user's preference in each snapshot is obtained,and in the second round,the user's preference in each snapshot is integrated to obtain the final preference.(2)Second,after getting user preferences,a user clustering algorithm based on Unsupervised Extreme Learning Machine(US-ELM)was proposed.The user's clustering result is obtained by inputting the user's final preference data into the US-ELM.(3)Third,two recommendation methods based on user preferences are proposed.One is a recommendation algorithm for user grouping,which uses the new item scoring algorithm to score the items of the group users,generates a recommendation list through the item's score,the other is a recommendation algorithm for a specific user by using a specific user's preference.Correcting the score of an item can make the recommendation more accurate.(4)Finally,experimental tests are conducted on real datasets and synthetic datasets,and compared with a variety of algorithms from the aspects of recommendation accuracy,recall rate,and time consumption.Experimental results show that the proposed algorithm has higher performance.