Research On Key Technologies Of Network And Content Analysis For Social Media

In the past decade,with the rapid development of Internet and web technolo-gies,social media has been involved in most aspects of modern life,instead of only acting as a communication tool.Social media consists of massive data per-taining to various application scenarios.It provides new perspectives and tools for studying the society,which is the advantage that the traditional sociology does not have.To better utilize and analyze these data,a great number of new data mining techniques and methods emerge focusing on social media.As the core aspects of social media analytics,the social network and content analysis have attracted plenty of studies.These work also results in many distinguished theoretical and technical achievements.In these two aspects,we investigate and explore the key technologies combining social media and corresponding applica-tion scenario.In this thesis,our contributions are highlighted as follows.1.We propose a hierarchical social network representation model based on hypergraph.It can satisfy most modeling requirements of different social networks by the component configuration.Then,we apply the proposed model on Facebook and another enterprise social network.In these two case studies,we analyze the characteristics of the network structure and validate some sociological theories,such as power law and small world effect.2.To handle the issue of node ranking in social network,we present a novel algorithm for bipartite graph inspired by PageRank and HITS.It supports two-type node ranking in the bipartite graph harnessing different metrics of node importance.This algorithm performs pretty well on the enterprise ranking problem.Through different evaluations,we find that it can rank the enterprise more accurately than baseline methods.3.Most link prediction methods in social network suffer from the huge amount of negative samples in the training set,which directly slow down the train-ing process.We present a new link prediction model based on network formation games.It's able to speed up the training while keeping the pre-diction accuracy by reducing the negative samples with game theories.The experimental results on various social network datasets demonstrate the effectiveness and efficiency of this model.4.With regard to the event feature extraction in social media,we propose an event-based topic model.It can not only extract the main aspects of the event but also capture the features of event category.Based on the extract-ed features,we analyze the events incorporating both traditional news and social media.In addition,we present an event classification approach utiliz-ing the extracted features,which indicates the effectiveness of the proposed model.5.In order to improve the effectiveness of unreliable content classification and better understand the content characteristics,we propose two classifiers based on logistic regression(LR)and deep learning respectively,as well as a new taxonomy of unreliable content.Firstly,we conduct the exper-iment on fake news detection,the LR and deep learning methods exhibit discriminate performance on different classification tasks.For the better accountability of the LR method,we analyze the linguistic style,sentiment,and subjectivity of fake news based on the extracted features from the LR classifier.Secondly,the proposed new taxonomy maps the intents with the reliability ratings of unreliable content.Then,we use LR and deep learning methods to classify the unreliable content under the new taxonomy.Final-ly,some interesting findings and patterns are obtained from the analysis of the unreliable content and the social reactions.In summary,this thesis investigates the key technologies of network and con-tent analysis for social media.It has theoretical significance and practical value for the study on data mining and analytics on social media.The proposed new models and methods can be applied in the real systems to achieve more efficient and accurate performance.The explorations on events and fake news in social media,provide the reference to better understand and solve these issues.