Application Research On Data Mining Algorithm In Mobile APP Recommendation

With the popularity of intelligent devices,a large number of mobile application markets have emerged,at the same time more and more mobile applications have been developed by developers.According to statistics,the number of applications installed by global mobile users in the App Store and Google Play in 2019 exceeded 120 billion times,an increase of 5% compared to last year.Massive applications bring users personalized experience and convenience,but also bring about the problem of information overload.Therefore,how to recommend applications for users to meet their individual needs has become a topic of common concern in academia and business.Nowadays,personalized recommendation technology has been widely used in areas such as business,entertainment and decision support.Traditional recommendation methods often use Collaborative Filtering(CF)algorithm to predict the missing values of the user-item rating matrix through similarity calculation or machine learning modeling.However,this method has serious data sparsity problems,so it is difficult to find similar users or similar items,resulting in poor prediction accuracy.In addition,compared with the traditional recommendation system,mobile APP recommendation has more new features such as: the variety of mobile applications and the rapid update,resulting in a more complex recommendation process than the traditional recommendation.Traditional recommendation often uses the user's overall evaluation of the item to obtain the user's preference for the item,ignoring the functional features and internal connections of the mobile application.However,in the download and use of APP,users' interests may be more inclined to their functional requirements.These problems lead to the traditional recommendation method directly applied to the mobile APP field has certain limitations.Therefore,this paper proposes two mobile application recommendation methods based on data mining.The main research contents and innovations are as follows:(1)An APP recommendation method based on the fusion of Hypertext Induced Topic Search(HITS)algorithm and association rules is proposed.This method combines the modified HITS algorithm with the association rules to achieve Top-k recommendation for the target users.The novelty of the method mainly includes: 1)In the iterative process of HITS algorithm,the user rating matrix was introduced and personalized information was added.2)By defining the user experience value,the influence of malicious users' extreme data on the recommendation results can be weakened.3)The weight factor is added in the traditional association rules,setting the importance of item set and the reliability of transaction,and redefine the weighted calculation formula of support and confidence,so as to reduce the redundancy of data and improve the mining efficiency of association rules.(2)An APP recommendation method based on functional features is proposed.This method implements personalized recommendation from the functional level of mobile applications by constructing the functional feature directed graph of user behavior.It not only considers users' demands for application functions,but also considers users' evaluation information,integrating user explicit feedback personalized into PageRank algorithm to obtain the user's preference for mobile application function.Based on these,a method for predicting the interest of the target user's APP is presented.(3)In order to verify the reliability of this method,experiments were conducted on the real data set of Huawei application market and compared with other methods.The experimental results show that this method is superior to the comparison method in the recommendation indexes.