Research And Implementation Of Automatic Selection Framework For Recommendation Algorithms

With the advent of the Internet,people can access all kinds of information without leaving their home,and the development of the Internet has shown an explosive trend over the years.According to the Statistical Report on the Development of Internet in China(hereinafter referred to as the Report),the number of Internet users in China has reached 829 million by December2018,and the number of new Internet users in the whole year has reached 56.53 million in total,which increases 3.8% over the end of 2017,and the Internet penetration rate has reached 59.6%.The explosive development of the Internet not only brings people convenience,but also serious data overload problem.In order to solve this problem,people have developed two solutions successively,namely,Classified Directory and Search Engine,but both of them have some defects.Websites with classified catalogue such as Yahoo and 2345,its catalogue can only cover a small number of items rather than long-tailed projects,and manual classification is too socialized and lack of personalization.For search engines,users can enter keywords in the search box,and the engine can find the desired information for users according to a certain algorithm and present it to users in a certain order.Although the scheme is fully personalized,the search engine is helpless if users do not or can not provide clear requirements.With the development of society,the Recommendation System also appears as a mainstream solution.The Recommendation System can analyze users' preferences based on their characteristics and recommend things they may be interested in or need.This “thing” can be a commodity or a piece of news.However,the current recommendation system usually targets recommendation software for a certain business implementation.When the business requirements or environment changes,the results obtained by the recommendation system will be significantly different.Therefore,this paper proposes a cross-platform general recommendation system framework with controllable algorithm,in which multiple optimized recommendation algorithms can be run,and the output results of each algorithm can be integrated and de-duplicated to achieve more accurate personalized recommendation.