Research And Implementation On Advertising Recommendation Algorithm Based On Fusion Model

Along with the Internet data's explosive growth,the amount and kinds of information grow more and more,people are facing the problem of "information overload".The search engine which help users to get useful information has been unable to meet the needs of personalized view of Internet users who have different characteristics.Being the branch in the research field of the individuation,the recommendation system can find the vast amounts of data which is relevant to the user information to recommend.In order to balance the contradiction between information and access to information,"information overload" problem also can get a better solution.This article obtains from the advertising recommended problems,this paper introduces the calculation of advertising related concepts.Then make recommend into CTR problems,build CTR algorithm framework to solve the problem of recommend.At first,the author first use traditional recommendation algorithm applied in advertising recommended scenario and recommended for traditional model shown on the ads feature set,considering the limitations of single model,put forward the model fusion method was applied to advertising recommended.Putting forward ideas and methods used in advertisements in the fusion of recommended,so we are going to use gradient to enhance the decision tree model(GBDT)and support vector regression(SVR)model integration.In order to make advertising recommended more accurately,we selected the data whose characteristics of the expressive force is more obvious as the basis of research.In this paper,several kinds of common traditional recommended method and fusion model is summarized,and its application is in advertising recommended scenario,the author analyzes the advantages and shortcomings of this a few class methods.Then the author found that the(GBDT+SVR)fusion model has the advantage to solve the problems of data sparse,subjectivity of feature selection,poor timeliness and so on.