Research On Weighted Slope One Algorithm Based On Likelihood Ration Similarity And Genre Correlation

The explosion of information scale in the Internet,meets the needs of the users of information.But the large amount of information makes it hard for users to locate the useful information quickly,leads to the emergency of information overload.Personalized recommendation technology is a kind of user-oriented effective means of personalized recommendation,which core is the recommendation algorithm.Collaborative filtering recommendation algorithm is popular because of its high prediction accuracy,and is not restricted by the type of project.In this paper,the classical Slope One algorithm is studied,due to its simplicity and high efficiency,the algorithm has been widely concerned.However,the algorithm lacks the consideration of user and item similarity,and the memory consumption is large,and the prediction results are not obvious compared with the traditional collaborative filtering algorithm.In order to solve the above problems,this paper improves the similarity calculation method and algorithm processing,and presents weighted slope one algorithm based on likelihood ration similarity and genre correlation.The main work is accomplished in this paper as follows:First,due to the calculation of the slope one algorithm did not consider possible correlation between users,the user similarity joins the algorithm processing procedure in this paper.However,the traditional users similarity calculation methods exist a lot of inherent limitations,so the likelihood ratio similarity calculation method is given in this paper.The idea to use a likelihood ration for similarity computations was inspired in part by the thought that linkage inheritance on the chromosome,and combined with the characteristic of the recommendation system data,improving the defects of traditional similarity calculation method,improving the quality of the screening of the similar neighbor users,providing a good data for prediction and recommended.Secondly,in view of the existing similarity measure method of the items did not consider the influence of genres of items,this paper suggests that use the genre of items to measure the relationship between items.This method measures the similarity of items throught the genres that the items belongs to and the ratings of items.What's more,considering that the slope one algorithm uses ratings deviation to predict,a item selection strategy is proposed.This strategy stabilizes the score difference,gets a local intensive scoring matrix and provides enough pre-processing for the algorithm.Finally,the experimental results show that the proposed algorithm has higher accuracy,compared with the current popular recommendation algorithm.This paper mainly has three parts:Firstly,creates a review by collecting and reading related documents.Secondly,it proposes the calculation method of user likelihood ratio similarity,and provides the calculation method of genre correlation of items and the item selection strategy,with which the performance of the algorithm will be further improved.Finally,it makes experiment analysis and performance assessment for the proposed algorithm,and makes summary and prospect.