AppUsage2Vec:Modeling Users' Usage Behaviors And Applications

With the rapid development of intelligent hardware in recent years,it has become an essential part of our daily life.As an important smart device,smartphones are changing humans' life style.Smartphones serve a wide variety of functions,and users can install and use mobile applications to achieve many imaginable purposes.The mobile application market has seen explosive growth in recent years.Mobile applications(abbr.apps)in a smartphone serve as entries for users to access a wide range of services.Frequent use of apps generates massive personal history data in app usage records.Since a smartphone is linked to an individual user,its app usage records can effectively capture large amounts of personal information,such as app usage pattern and intent.This work attempt to understand users' app usage behaviors by analyzing app usage records.Understanding users' app usage behaviors have several implications.First,it can help mobile phone manufacturer optimize the operation systems so as to speed up mobile phones.Second,knowing users' app usage pattern can be helpful for recommending apps to users depending on their habits.Finally,it can also help users understand their app usage behaviors objectively and extensively,to break bad habits so as to improve life quality.This work attempt to understand users' app usage behaviors by proposing an AppUsage2Vec model based on app usage records,with which users' app usage behaviors are modeled and understood.The work can be summarized as follows:1.Proposing an AppUsage2Vec deep model to model the behavior of App usage,and modeling users' app usage behaviors.The model is improved by considering the correlation between users and apps,and the relationship among apps.2.Using the AppUsage2Vec model to predict what next app will be used.Our approach is evaluated with a dataset of app usage behavior records of about 100,000 users.When we use 5 apps to model the behaviors,the accuracy of app usage prediction for 1 candidate app,5 candidate apps,10 candidate apps,and 15 candidate apps is 48.24%74.86%,84.71%,and 90.54%,respectively.Compared with the methods in the literature,our approach is better by up to 10%..3.Applying the AppUsage2Vec model to detect the co-used apps.Employing the embeddings learned by AppUsage2Vec model,this work clusters the apps to discover app clusters in each of which the apps are co-used.Moreover,the users are clustered to different user groups,where the users co-use similar apps.