An Intelligent Recommendation Algorithm Based On The Multi-dimensional Regression And Collaborative Filtering

The rapid development of science not only promotes the economic and socialprogress,but also brings convenience to people’s life.The technology of Web2.0becomes mature and further promote the development of e-commerce.Currently, the number of e-commerce platforms based on internet is increasing.And its applications are increasingly widespread. It leads to the increasing amount ofinformation. The increase of information quantity will give users more options,however it led to user’s information overload, unable to extract or find the informationthey want. Therefore the personalized recommendation system based on the supplierof user and the electricity was born.For the user, a superior performance of therecommendation system can take the initiative to recommend products to users whatthey interest in, not only saves time, but also give users a nice personalized experience.For the electricity supplier, recommendation system has enormous commercialpotential. It can maintain the old user groups, at the same time it can identify morenew user groups.As a core part of the recommendation system, personalized recommendationalgorithm has a direct impact on the user satisfaction and electricity supplier. In viewof this, recommendation algorithm has become a star in today’s e-commerce. In orderto provide users with more efficient and more accurate recommendation service, andbring greater business value to electricity. People improve existing algorithmssimultaneously, and also continues to study the new algorithm constantly. In this paper,firstly, we analyze the deficiency of the existing recommendation algorithms.Secondly, we improve the algorithm from two directions which are user interactiveand no-user interaction products. Finally, by the experiment of simulation, we haveverified the effectiveness of the improved algorithm in this paper. Content of thispaper is as follows. （1）Forecast the user interactive products. In this paper, we present amulti-dimensional logistic regression recommendation algorithm which is based onthe traditional logistic regression; it introduced the characteristics of the system timeand Bootstrap re-sampling techniques. Due to the introduction of Bootstrapre-sampling technique, it led to some problems of parameter settings, such as numberof sub-models, number of features, positive and negative samples. In order to solvethis problem, we use the genetic algorithm for parameter optimization. The fifthchapter of the simulation result shows that the improved algorithm has a good effectfor the forecast of interactive products.（2）Forecast the no-user interactive products. In this paper, we present acollaborative filtering recommendation algorithm. Before user-product rating matrixhas generated. Firstly, we use the memory oblivion function to attenuate, thencalculate the similarity and consider the user attributes and product attributes. And weuse logistic regression to train the feature weight.Finally, using an appropriate weightto combine the results of collaborative filtering recommendation algorithm which isbased on user and project. Based on the above improvements, the improvedcollaborative filtering recommendation algorithm in this paper has got a bigpromotion of prediction accuracy.（3）Finally, we combined two prediction algorithms, user interactive productsand no-user interactive products, effectively. The prediction performance has beenimproved further.