Research On Recommendation System Of Operators' Platform Based On Hadoop Technology

As an intelligent and personalized information service system in the operator platform,the recommendation system can analyze the user's personalized demand information and establish a recommendation model,formulate reasonable and efficient recommendation strategies.how to find and recommend products to meet the user's needs in the massive operator platform,improve the user experience,improve the viscosity of the platform,and become the current major operators Problems to be solved in the context of "speeding up and reducing fees".At present,all major operators are launching the "mobile phone binding package" marketing model through their own electronic channels(such as online business halls),the purpose is to try to find new business growth points.In fact,the proportion of users who access or purchase mobile phones through online business offices is still low,resulting in a sparse user-product rating matrix in the background.However,the traditional recommendation algorithm has poor recommendation effect when processing implicit feedback information of sparse data users,and cannot meet the real-time problem of the operator platform recommendation system.Therefore,this paper focuses on improving the recommendation efficiency and the real-time performance of the recommendation system,and designs a mobile phone sales recommendation system in line with the characteristics of operators.The main work and achievements are as follows:1.In view of the low accuracy of the traditional recommendation model when recommending item resources on the operator platform,this paper adds a dual clustering technology that can mine local information relationships to build a recommendation algorithm model based on the original calculation of user or item similarity.According to the MATLAB simulation experiment,it can be seen that the accuracy and recall rate of biclustering technology are improved by 22.6% and 29.1% respectively compared with other traditional algorithms,which can ensure users to get correct recommendation results in time and efficiently.2.Aiming at the problem that traditional recommendation can only be processed offline,this paper designs a real-time recommendation system architecture based on Hadoop and stream processing technology,store recommendation results in a distributed cache.Then introduce the Storm architecture to perform real-time recommendations based on feedback and requests from users' current behavior,in order to avoid the situation that a single Hadoop architecture causes the recommendation process to stay at the previous stage,and make the generated recommendation list more reasonable to meet the needs of users.The problem of cold boot is still a problem that the operator platform faces in the recommendation system.At the same time,the user's interest may change with the change of time and environment.How to explore more user's interest has become a difficult and key problem.It will use trend recommendation in the follow-up work,combine the time characteristics of scoring,and make further research and discussion on this problem.