Research On Recommendation Algorithm Of Mobile Application

With the popularity of smart phones, the number of phone application rapid greatly. According to statistics, the number of applications in android and iOS software store both exceeded one million in 2014. It costs lot of energy for user to find application that they are interested from mass applications. Mobile applications store has its own review system as same as traditional products, that users can review the application easily after downloading. When looking for a new application, user will view application info, another major factor is other user’s comments reference. Faced with huge amount of applications, the user cannot see all the information to get the best one. Therefore, an effective recommendation system is required to solve the info-overload problem faced by users.Traditional recommendation algorithm, such as collaborative filtering system, consider the product as static unit, which mean that the product properties will not change over time. However, mobile applications are different from traditional products. The biggest difference between mobile applications and traditional products is that properties of mobile applications will vary with the new version. To recommend applications for users, the change should be taken into account. After concerning the dynamic features of mobile applications, the accuracy of mobile applications recommendation should improve obviously.We consider about the dynamic features of mobile applications and competitive scores between similar applications, then we suggest a new mobile application recommendation algorithm. The algorithm aims to reflect the dynamic features of the recommended application, and recommend a competitive application for user from similar app list. The model uses LDA method to analyze the dynamic features of phone applications, and then make these dynamic features properties of app. When collecting text used for LDA model, we use application description text. What’s more, we also add the version upgrade logs into the text, which should help analyze the dynamic features. When generating a recommendation list, we consider competitive score of between similar applications. When faced with lots of similar apps, user will compare the new app with used app, and then choose the best one to download.We consider the different features between apps and traditional products, which is dynamic changes caused by app updating. We add updating logs of apps into text used for LDA model so that the result of LDA can include the dynamic features. Meanwhile, we filter the competitive apps by competition mechanism, expecting that will improve the accuracy of mobile applications recommendation.