The Research And Application Of Personalized Recommendation System Based On Combination Algorithm

By the advent of the Network and the rapid development of Internet technology, it hasgradually changed peoples’ traditional operation mode of information resources. It’s verysimple, fast and efficient for disseminating information in the network, the total amount ofwhich explosively increase. The phenomenon of Information overload will emerge becausethe value information may be submerged in the ocean of information when user browsesinformation in the internet. In this background, the personalized recommendation system isproposed, but it has some problems in the current personalized recommendation systems. Forexample, recommended strategy is single, recommended automation is low and lack ofpersonal issues. Therefore, it has become a hot research spot that how to design an effectivepersonalized recommendation system. This paper deeply do research on the problem that howto improve users’ satisfaction of the recommendation. The main work of this paper is asbelow:1) This paper will propose a new recommendation algorithm against the problems of thecurrent popular recommendation system such as sparsity, cold start and accuracy. First of all,it will improve cold-start problem in Slope One algorithm by recording the user clicks.Secondly, the context probability is introduced to calculate the similarity and real-timeincremental to update the similarity matrix. Finally, the efficiency of the recommendation isenhanced when the matrix is sparse though introducing the content information and similaritycalculation based on content and evaluation matrix The result shows that the combinedalgorithm can solve the problem of sparsity, and real-timely reflect the changes of userinteresting.2) The user interesting model and the data item model for adapting the personalizedrecommendation system is proposed. The user interesting model is established according tothe laws of the human brain memory. It uses the users’ interesting model based on the laws ofhuman brain memory to simulate the changes of the users’ interesting. And the quantificationmethod is explited to display the users’ interesting. The data item model refers to transform adata item into the representation process of system. Such as data items are directly expressedas word segmentation. The similarity comparison resulted by this data item model may spenda lot of time and may lose some useful information. This paper makes use of FLD (FisherLinear Discriminate) algorithm to reduce the dimension of project feature space, and use theresult to establish project topic model, so it is a better method to solute the over-fittingphenomenon.3) Based on combination recommendation algorithm, user interesting model and the dataitem model, it achieves a personalized recommendation system prototype called ECRS which aims at book channel of e-commerce sites. It uses a flexible recommendation engine, whichhas a recommended strategy framework combining and coordinating different kinds of therecommendation algorithms, and also choosing the right algorithm for different scenarios toimplement recommended tasks. The data storage interface, the recommended interface andevaluation interface in the system can be well adapt to different applications.