An Improved Long-tail Recommendation Algorithm Based On Item Collaborative Filtering

At present,the traditional recommendation system has serious long-tail effect.Recommender systems often only recommend current popular items to users,while items with higher value but less popular items are not well recommended.Therefore,in order to solve the above problems,this paper proposes an improved long-tail recommendation algorithm based on item collaborative filtering.The main work of this paper includes the following:(1)The current item-based collaborative filtering recommendation algorithm has serious sparsity in the data set.The sparsity affects the calculation of similarity.The sparser the calculation,the lower the recommendation accuracy.Get better digs.In order to solve the sparsity problem,the K-means clustering algorithm is introduced,which achieves the effect of clustering by randomly selecting the initial seed and continuously optimizing and iterating.This paper improves the traditional K-means clustering algorithm and proposes a K-means clustering algorithm that integrates item popularity.This method introduces the concept of popularity,calculates the popularity of each item,selects the initial seed through the popularity,and continuously optimizes and iterates.Experiments show that the improved clustering algorithm further improves the mining ability of long-tailed items,and has better search ability for long-tailed items.(2)Comparing the traditional item-based collaborative filtering recommendation algorithm with the user-based collaborative filtering recommendation algorithm,it can be found that the item-based collaborative filtering recommendation algorithm is rich in longtail items,and can be greatly excavated for the personalized needs of users.The value of long-tail items.The long tail factor is cited in this paper,which takes into account not only users but also items.When calculating the similarity,the weight of popular items is reduced,the weight of unpopular items is increased,the weight of active users is reduced,and the weight of ordinary users is increased.Two indicators of coverage and popularity are calculated,and the experimental results show that this method can effectively mine unpopular items.(3)After obtaining the similarity,when recommending to users,Top-N recommendation is used.The traditional Top-N recommendation recommends the top N items with the highest similarity to the user.This method can quickly obtain the items closest to their own interests for the user,and has a good accuracy rate,but the coverage of this method recommends The rate effect is not very obvious.This paper improves on this basis and introduces a frequent scaling factor by which the average popularity of the recommendation list is adjusted.The experimental results show that this method can effectively improve the coverage of the recommendation and reduce the popularity of the recommendation sequence.