Research Of Social Network Information Propagation Model Based On Multi-dimensional Attributes

With the rapid development of social media platforms (Social Network Site), such asFacebook, Twitter, Youtube, WhatApp, Microblogs, Weixin etc., social media analysis hasbecome a very popular research area. Information propagation, as one of the most importantcharacteristics of social media, is a major topic of social media analysis and has received agreat deal of attention from researchers. Studying the model and patterns of informationpropagation has a very important theoretical significance. It can help us analyze the attributesof network nodes, understand the functions of networks, reveal the flow of information, andpredict the width and depth of information spreading. Meanwhile, studying informationpropagation in social media has a wide range of application prospects. It is being applied tomany fields, such as public sentiment monitoring and analysis, rumors tracking, businessadvertising, and personalized social marketing.This thesis aims to study the information propagation in social media. As to the datapreparation, developed a parallel and efficient system to crawl social media data. Takingstatic attributes (networking attributes and non-networking) and dynamic attributes intoconsideration, we propose a method to analyze the relationships of community structure andlocation information and give four Trajectory Patterns extracting algorithm. We proposeInfo-Cluster to describe how the information originated from a location cluster propagates inor between communities, and then we present the information propagation based Info-Clusterdetection algorithm. With the networking attributes (social relationships) and non-networkingattributes (self behaviors), we propose a new social network data model named Multi-LayerNetwork (MLN) and present a new information propagation model. Considering the dynamicattributes of social networks, we present a N-Pat Tree model and give some filteringmechanisms to detect new popular phrases multi-dimensionally. We also propose a networkmodeling method to integrate the dynamic attributes and give a new information propagationmodel. The main contributions of this thesis are summarized as follows.（1）Large scale social media data obtaining and directed analysisData preparation is the basis of the research of social media analysis and mining.Compared with the traditional data, social media data could be characterized as large scale,complex structure, rich content, and flexible obtaining. In this thesis, we have developed aparallel and efficient system to crawl social media data. Then we perform some analysis onthe attributes of the nodes in social media data. The main work of this part can be summarizedas follows. Considering each individual’s posting frequency and staying time on a certain location, we define a main location of an individual. In order to study the relations betweencommunities and location clusters, we propose the index of location entropy to measure thedegree of dispersion of the locations in each community, and the index of community entropyto measure the degree of dispersion of the communities in each location cluster. At last, weanalyze users’ trajectories and define four Trajectory Patterns. An algorithm is proposed toextract those patterns from microblog data. We also apply the proposed algorithm in thediscovery of the relatives of cities and hot cities. The above work has made a good data andtheoretical foundation for the following research on information propagation.（2）Information propagation analysis based on location information and communitystructureThe existing works on information propagation mainly focus on the analysis of thenetwork topology. However, they ignore the influence of nodes’ attributes or nodes’ states oninformation spreading. For instance, the location and community information often implyabundant information and determines the individual’s spreading behaviors to some extent. Tosolve this problem, we propose Info-Cluster, an innovative concept to describe how theinformation originated from a location cluster propagates in or between communities, andthen we present the information propagation based Info-Cluster detection algorithm. The maincontributions of this part can be summarized as follows. In order to satisfy the diversity ofinformation propagation, we take the location and community information into consideration,and give different spreading parameters for different attributes. In order to study the width ofinformation spreading qualitatively, we take the location information into our algorithm toensure the credibility of events. In order to study the width of information spreadingquantitatively, we take the community information into our algorithm to improve thecredibility of information propagation. Therefore, the proposed information propagationmodel based Info-Cluster detection has many potential effects on event monitoring andemergency processing.（3）Information propagation analysis over Multi-Layer NetworkSocial media platforms have become ubiquitous for social networking and informationsharing. People involved in those platforms often post some interesting information, addfriends, browse or comment on the postings of their friends. From this point, we can find thatevery individual of social media has networking attributes (social relationships) andnon-networking attributes (self behaviors). How to study the information propagation with theabove two kinds attributes has become an important problem. In this thesis, we propose a new social network data model named Multi-Layer Network (MLN) over microblog. In the model,different layers represent different kinds of relationships between individuals. We present anew information propagation model based on the MLN model. Finally, we conductexperiments on real-life microblog data of four recent hot topics. The experimental resultsshow that our MLN model and influence propagation model are more effective in finding newand accurate active individuals comparing with the single layer data model and the linearthreshold model.（4）Information propagation analysis on the social networks with dynamic attributesThe attributes of users in social media, including networking attributes (socialrelationships) and non-networking attributes (self behaviors), are often influenced by users’current behaviors, which results in the dynamic characteristics. To study the dynamicattributes and the corresponding information propagation model, we first extract key wordsfrom users’ postings, comments or annotations. Then we define a TF-IDF similarity matrix tocalculate the similarity of users. We propose a network modeling method to integrate thedynamic attributes and give a new information propagation model. During the key wordsextraction, we find that social media is generating more and more new popular phrases. Tosolve this problem, we present a N-Pat Tree model and give some filtering mechanisms. Wealso propose an algorithm to find and analyze new popular phrases multi-dimensionally. Theexperiments on one-year Tencent-Microblogs data have demonstrated the effectiveness of ourwork and shown us many meaningful results.