Research On Hybrid Recommendation Algorithm With Stability Of Interest And High Utility Factor

With the rapid development of information technology,the phenomenon of information overload has become increasingly urgent.In massive information,it is difficult for users to portray what they need and are interested in.To help users solve this problem,both the academic and industrial sectors,the recommendation system has been becoming hot.The core of the recommendation system is the recommendation algorithm.Meanwhile,with the development of the Internet and the rise of artificial intelligence,in recent years,various recommendation algorithms have emerged in an endless stream and have achieved remarkable results,but there are also some problems,such as accuracy problems,long-tailed phenomenon,popularity bias and so on.This paper mainly focuses on improving the accuracy of recommendation,mining long tail items,alleviating the phenomenon of popularity bias and strengthening the quality of TopN recommendation.including:(1)In order to fully capturing user interest,starting from the nature of user interests,taking into account the user's interest stability as well as focusing on time-sensitive,defines the user interest and the user's interest vectors for the item.Finally,a similarity model with stability of interest and time-sensitive is proposed.(2)In order to fully mining long-tailed products and mitigating long-tailed phenomena,considering that the popularity of the items will also have an impact on user interests,Therefore,this paper builds interest modeling for user by using the popularity of items and proposes an user interest feature similarity model with item popularity.(3)In order to better grasp the user's interest and an accurate recommendation to the user and a good recommendation experience to the user while mitigating the popularity bias of the item,between the feature similarity model with the popularity of items and the time-sensitive similarity model with interest stability are linearly weighted,under certain condition that without increasing time complexity of the algorithm,exponentially.and then a hybrid recommendation algorithm with stability of interest and time-sensitive is proposed.(4)In order to fully consider utility of the user and the potential behavior of user,which is divided into two phases: user rating and item selection.Firstly,the topic model is used to mine the latent high utility factor and targeted probabilistic factor.and then the two factors are weighted fusion as the first stage,linearly.In order to better predict user specific ratings,the second stage uses singular value decomposition model.Finally,ranking value of user interest is basis by integrating two stages.Above all,a hybrid recommendation algorithm of weighting high utility factor based on two-phase is proposed.The experimental results show that a hybrid recommendation algorithm with stability of interest and time-sensitive can excavate long-tailed items as well as improving the accuracy of the recommendation algorithm,and mitigates phenomenon of the popularity bias of item.In addition,a hybrid recommendation algorithm of weighting high utility factor based on two-phase improves the quality of TopN in large-scale data sets meanwhile,availability and scalability of algorithm is tested and verified.