Research On Key Issues Of Information Diffusion In Social Networks

With the rapid development of Web2.0 and the popularization of mobile intelligent terminal,various types of social media and social networks emerge and prosper,which is profoundly changing the way people work,live and communicate.The information communication based on online social network presents many new features,including the large audience,the speed of information dissemination,the complexity of network structure,the wide influence range,and the dynamic evolution of the network.To further study the law of social network information dissemination is helpful for us to understand the mechanism of network evolution,predict the trend of information dissemination,which has important commercial value for enterprises to develop marketing strategy,brand promotion and online advertising.For the government and public security departments,the research on the information dissemination in the social network can help to grasp the trend of public opinion,control the network rumors,and protect the national security.Information diffusion in social networks is a complex dynamic process,which usually involves many factors,including the structure of the diffusion network,information content,communication time,user attributes,etc.These factors are interdependent and closely coupled,which has brought many challenges to the research of information diffusion.The challenges in this area include the complexity of network structure,noise and informal information content,the complexity of diffusion phenomenon and the difficulty when tracking information diffusion pathway.The research of information diffusion in social networks involves many aspects,and the key issues include: information diffusion mechanism or model,information diffusion network structure,information diffusion maximization and information diffusion outbreak detection,etc.In this paper,we mainly focus on these core issues and give some in-depth investigates.In detail,we highlight the main contributions of this thesis as follows.Firstly,we analysis the information diffusion mechanism in social networks from micro perspective and propose a fine-grained information diffusion model based on node attributes and information content.Traditional models pay more attention on network structures,and largely ignore important dimensions such as user attributes or information content features.In this thesis,we extract features from multiple dimensions including node attributes and information contents,proposing a fine-grained information diffusion model based on these features.We use stochastic gradient descent method to learn the weights of these features in the proposed model,and then give a quantitative analysis.Our model gives an insight into the extent on which these features can influence information diffusion on social network.Besides,given an initial state of information diffusion,our model can be used to predict the future diffusion process.Experiment results on Sina Weibo dataset show that the proposed model yields a higher prediction precision than other widely used models like AsIC model or NetRate model.Secondly,we study the problem of information diffusion network inferring and pathway tracking and propose a novel probabilistic model called Network Inferring from Multidimensional Features of Cascades(NIMFC)?In real-world,many times we can observe the trace of contagion spreading across network,but the underlying network structure is unknown and the transmission rates between node pairs are unclear to us.In this paper,given the observed information cascades,we aim to address two problems: diffusion network structure inferring and information diffusion pathways tracking.We propose a novel probabilistic model NIMFC which takes into account heterogeneous features,including temporal and topological features of cascades,node attributes,and information content,to infer the latent network structure and transmission rates of edges.Also,based on the inferred network structure,we may track diffusion pathways of a cascade in social networks.Our proposed model NIMFC is evaluated both on large synthetic and real-world data sets,and experimental results show that our method significantly outperforms stateof-the-art models both in terms of recovering the latent network structure and information pathway tracking.Thirdly,for information spread maximization,the traditional greedy algorithm has disadvantage of low efficiency.To address this problem,we attempt harnessing historical information cascades data to propose a data-driven approach named GA-LIM(Greedy Algorithm based on Local Influence Model)and a topic-sensitive information spread maximization algorithm named Topic-Max.For the GA-LIM algorithm,we first propose a voting algorithm to learn diffusion probabilities of edges from cascades data.Then a pruning method is developed to remove trivial edges whose weights are smaller than a threshold.Moreover,motivated by the social influence locality,we propose a Local Influence Model to evaluate node's influence within a local area instead of the whole network,which can effectively reduce the computational complexity.Based on local influence model,we use greedy algorithm to find an approximate optimal solution.For the topic-sensitive information spread maximization,we take into account the topical feature of information content and use topic model to learn a topic mixture for each propagation message.Based on this,we build directed and weighted diffusion network for each topic.Last,we apply greedy algorithm to identify influential nodes with respect to a specific topic.Lastly,we study the information diffusion detection strategy in social networks.Incorporating network structure,node attribute,historical information cascades and detection cost,we propose a random-walk based algorithm DiffRank to sort nodes according to their diffusion ability,then we choose the top-k nodes on the list to place sensors and detect information diffusion.Experiments on real dataset of Sina Weibo show that DiffRank outperforms state-of-art algorithms with respect to information cascades coverage ratio,detection time and reduction of infected population.Besides,DiffRank can be implemented easily in distributed or parallel computing environment,achieving good scalability.