Research On Item Collaborative Filtering Recommendation Algorithms Based On Long Tail Theory

Traditionally,Pareto's law seems to think that the company's large income is generated by relatively few hot products,so the previous e-commerce sales data is more concerned with the top products.However,the development of the Internet has overcome the difficulties of traditional retailers.In the case where the cost of goods at the edge of the commodity tends to zero,the sales of those products distributed in the tail are sufficient to match the hot goods.Through the analysis of sales data of Internet companies such as Amazon and Netflix,Chris Anderson,a magazine editor in the United States,proposed the concept of "long tail",which has aroused widespread concern.Researchers in various scientific fields have existed in life.The long tail phenomenon has been studied in widespread.This paper combines the relevant concepts of long tail theory.In order to improve the recommendation rate of unpopular products,the article improves the TOP-N algorithm based on item-based collaborative filtering.The specific work is as follows:Firstly,this paper introduce the main algorithms in the recommendation system field.While analyzing the advantages and disadvantages of different types of recommendation algorithms,it focuses on the difference between the item recommendation itemCF and userCF based recommendation in the collaborative filtering algorithm.The three perspectives of data sparsity and diversity to explain why the item-based itemCF is used as the basis for the long tail theory recommendation.Secondly,for the traditional long tail recomnendation based on item recommendation,the unpopular goods and the popular goods are not considered in the calculation of the similarity of the items.This paper proposes an improved algorithm itemCF-IIF based on the item weighting factor,which punishes the frequency at popular items appear on the recommendation list,increases the probability that unpopular items are recommended and compares them with other similarity improvement algorithms.Thirdly,in order to solve the problem that Jaccard's similarity algorithm can not differentiate the item scores,an average score factor is proposed to rank and optimize the items in the user,s recommendation list.The user recommendation list optimized by ranking improves user experience and recommendation quality while mining long tail distribution items.Finally,in view of the drawbacks of swiping scores phenomenon in long tail recommendation,this paper analyses the influence of average score of user evaluation on experimental evaluation standards through experiments,eliminates the most suspected users and improves the accuracy of recommendation results.Figure 19 table 10 reference 57...