The Recommendation System Of Long-tail Product

With the development of the Internet and technology,the volume of information in various aspects of the public has exploded.In particular,the development of mobile phones motivates everyone to catch the mobile phone,such as a shopping platform on the mobile phone,so that users can go online and shopping anywhere at anytime.Of course,a lot of data has also been produced.Therefore,it is especially important to find valuable data in a large number of sparse data on the e-commerce platform.The recommendation system has very significant advantages for analyzing the data.The current recommendation system gradually matures,but these systems are focused on resolving recommendations for popular products,while ignoring the recommendation of Long-tail Product.The commodity types of the entire business platform are divided into popular products and long-tail product.Although there are few users,the total purchase quantity can be roughly equivalent to that of popular products.At the same time,because long-tail product cannot be directly displayed on the main recommendation page of the e-commerce platform,personalized recommendations for long-tail product are indispensable.First of all,this article gives a summary introduction of popular goods,including the concept of popular goods,the characteristics of popular products,and the characteristics that should be possessed in the recommendation system of popular products.Only analysis and research of popular products can make the improved recommendation system more targeted.Then,some recommender systems that are commonly used in electronic goods platforms are briefly introduced,especially collaborative filtering systems,and the advantages and disadvantages of different types of collaborative filtering algorithms are compared.Secondly,the cluster analysis is proposed to improve the classification and the article-based collaborative filtering algorithm.On the one hand,when the user's preference for items is calculated,the user can use the invisible feedback information reasonably.Instead of treating invisible information of all types of users equally,it assigns different proportions to different types of information,reflecting the importance of user information types in calculating user preferences.On the other hand,the item similarity algorithm improves the algorithm.Eliminating the impact of hot items and maintaining the effectiveness of the algorithm does not result in infinite results for new users.Finally,using the real data of Alibaba's e-commerce platform,conduct case.Two evaluation indicators,coverage rate and accuracy,are used to analyze the results and evaluate the performance of the proposed algorithm.In this paper,the sparse user information and the existing recommendation algorithms for the popular product recommendation system do not take into accountThe impact of hot products on the unpopular products,so an recommendation system is proposed.The K-means clustering in the cluster analysis is used to divide the users in the general direction,thereby alleviating the sparseness of the user ratings matrix of the long-tail product.The advantage is that the K-means clustering operation is fast,and it has a good ability to cope with the rapid update of information in the e-commodity platform.At the same time,as the information is updated,the family center of the cluster can be continuously changed,and the scalability caused by the new user is relieved problem.The improved recommendation algorithm is based on the article-based collaborative filtering recommendation algorithm.Because users who often use e-commerce platforms often do not like the recommendations of mainstream products,and more often recommend items that they are interested in,article-based collaborative filtering algorithms are a good choice.When the item-based collaborative filtering algorithm calculates the degree of user's preference for a certain product,the user's invisible feedback information such as purchase,collection,plus shopping cart,and click is not equal to different information,but is calculated by using a comparative scale weighting method.The weights make the calculation results fit the actual situation more accurately and accurately estimate the user's preferences.At the same time improve the formula for calculating the similarity of items,try to eliminate the impact of popular products on the recommendation results,and also avoid the results that some new users of cold products cause the denominator of the formula to be zero.The coverage rate analyzes the recommendation result.Because the recommendation system given by the recommendation system with a high coverage rate has a large variety of products and these products have rarely been recommended.in the past,the recommendation system has a strong capability of excavating a popular product.The accuracy rate can indicate the difference between the prediction and the real situation,reflecting the accuracy of the recommendation system.Therefore,coverage and accuracy are used as indicators for evaluating the recommendation system.According to the real e-commerce data,the cross-validation was performed.The results of the obtained indicators showed that,although the accuracy rate was not much higher than other recommended systems,the coverage rate had a higher index.Since the coverage rate indicates the popularity of the products offered by the recommender system,the recommendation system of the present document has the ability to recommend cold goods better.