Mining Dynamic Heterogeneous Data With Distributed Algorithms

With the rapid development of Internet and smart devices,the volume of data being collected has reached an unprecedented scale.It is of great challenge to analyze and process such amount of data.Currently,how to effectively deal with these data is one of t he hottest research topics.Distributed computing has become a key in processing big data and many related problems require specific designs on distributed algorithm.In this work,we realize distributed models to maximize the influence of information propagation and to extract patterns of online user behaviors.The prevalence of online social network has tremendously accelerated the spread of i deas.In the first part of our research,we focus on t he problem of influence maximization in mobile social networks.The goal is to maximize the propagation range of information by selecting appropriate information spreaders.After preliminary analysis on the dynamic property of online networks,we establish the influence optimization problem under constant economic constraints.The study proposes a dynamic algorithm to firstly partition the network and select seeds according to real-time status of the propagation.Experiments show that our a lgorithm can efficiently maximize the influence under dynamic conditions.We also study the problem of inferring user interest by their online browsing sequences.The research is carried out on a real e-commerce dataset containing over a hundred million records.Finally,we propose a double-layered graphical model to describe the motivation behind these sequences of user behaviors.Experiments verify the effectiveness and the robustness of our algorithm.