Design And Implementation Of Recommendation System Based On User Portrait

In today's rapid economic development,the Internet and e-commerce become essential role in people's lives.On the one hand,it brings great convenience to the people;on the other hand,it generates expansion and redundancy of information.As a result,facing a great number of information,people cannot make a wise choice.How to provide information that is correct,requirement-met,and fast by using the simple and easy Information Retrieval(IR)becomes the top priority of the majority of professors.Therefore,the Recommendation System(RS)is introduced.On the basis of domestic and international successful cases and learning relevant references,this thesis designs a RS consistent to the logic of current services based on user portrait,which implements expected recommendations.After mixing with the recommended method set,this RS completed a more accurate recommendations.This RS is composed of four parts:building user portraits,portraits visualization,bayesian automatic grading and mixed recommended.In the first part of the construction of the user portrait stage,the system in the data extraction process using multi-dimensional collection,the use of real-time computing framework Storm to collect and process large amounts of data collected at the same time,then call the user behavior has been encapsulated predictive interface,the resulting data In order to meet the non-technical staff can fully understand each user's basic information,behavioral information and purchase information,using SpringMVC framework and Bootstrap,EasyUI and ECharts and other plug-ins.In the user portrait visualization,in order to meet the non-technical personnel to fully understand each user's basic information,behavioral information and purchase information,A new multi-dimensional dynamic user visualization platform is constructed.In the Bayesian automatic scoring part,based on the original recommendation strategy set,a commodity-based historical commentary is added to the product scoring strategy,and then mixed into the existing recommendation result The system at this stage the use of Bayesian theory,the principle of Bayesian classifier and Hadoop distributed framework for the analysis of the historical reviews of goods,and then on the merchandise In the mixed recommendation list stage,the recommendation result obtained by the original recommendation result set of the enterprise is used for mixing the recommendation result of the preference prediction recommendation and the automatic score according to the weight proportion,and the recommendation result of the recommendation result set of the enterprise is mixed according to the weight proportion,Get comprehensive merchandise recommendation list,displayed to the user.Grasp the data for a period of time,after testing the test concluded that:the application of this system,making the recommended product confidence,service and evaluation are excellent merchandise in the list of recommended products in front of the personalized recommendation with a clear,Resulting in more clicks,impressions,and purchases,which has brought more benefits to the enterprise and further validates the effectiveness of the proposed strategy.