社交网络舆情传播与控制研究

发布时间：2018-03-15 23:08

本文选题：社交网络　切入点：微博　出处：《哈尔滨工程大学》2016年博士论文　论文类型：学位论文

【摘要】：近年来,随着社交网络平台的发展,社交网络已经逐渐取代传统媒体,成为人们获取信息、传播信息的主要平台。社交网络中信息数据的产生传播也更为方便,在为信息传播提供便利的同时也为谣言,反动信息提供了更好的传播平台。与现实社会不同,社交网络的舆情产生可以不受时间、空间的限制,使得社交网络的舆情更加难以控制。当负面社交网络舆情大规模爆发时,如果不能及时有效控制负面舆情的发展,其将会成为影响广泛的社会事件,危及整个社会的安全。因此,面向社交网络舆情安全的研究已经得到学界和业界的重点关注。本文以最具代表性的社交网络平台微博作为研究对象,旨在研究社交网络舆情传播的影响、预测以及控制。由于微博网络具有数据量大、信息碎片化严重、交互多样性、信息传播快等特性,通过系统审核或人工实时监控等方式并不能有效地限制社交网络舆情危机信息的传播。因此,针对微博这一社交网络舆情主要推动媒介,如何自动实现微博舆情的传播影响度量、预测、对负面舆情信息的传播控制与监管,有效阻止微博舆情危机爆发并正确引导微博舆论是社交网络舆情安全保障中亟待解决的重要问题。本文从社交网络舆情安全的角度出发,主要针对以下几个方面开展研究:首先,针对大规模社交网络数据中存在大量碎片化数据,并且这些数据具有较高的稀疏性,提出一种面向多实体稀疏关系数据的高效联合聚类算法。在算法中,为了充分利用多关系数据,提出了一种鲁棒的约束信息嵌入方法构建关系矩阵,降低了矩阵的稀疏性,进一步提高了算法的准确率。在稀疏约束的块坐标下降框架下,关系矩阵通过非负矩阵三分解算法同时获得不同实体的聚类指示矩阵。非负矩阵分解过程中,通过高效的投射算法实现快速求解,确保了聚类结果的稀疏结构。在人工和真实数据集上的实验表明,算法在三个指标上都具有明显提高,特别在极端稀疏数据上的效果更加明显。其次,针对社交网络舆情事件相关消息传播影响度量问题,在综合考虑交互用户历史属性、用户间交互类型及交互频次的基础上,提出了基于用户交互的微博有向传播树模型。在该模型中建立用户交互度量矩阵来描述用户间的交互,对单消息传播过程中的用户交互图进行稀疏化处理形成有向传播树,来刻画消息快速传播过程描述。利用有向树中边的交互度量矩阵进行消息影响力度量和垃圾消息识别。在真实的新浪微博数据集上的实验表明,传播树模型能够有效地刻画消息传播过程,同时能够有效地度量消息的传播影响。再次,针对社交网络舆情事件相关消息传播预测问题,考虑关键用户对舆情事件传播的影响,提出了基于关键用户的动态线性预测模型。为了避免结构属性对消息传播规模预测的影响,将消息的传播规模刻画成多个用户的传播规模问题,并建立用户传播影响函数。为了提高线性模型的拟合效率,仅考虑关键用户的传播影响函数生成线性预测模型。利用预测值和实际值的差异检测舆情事件消息传播中的关键用户,并根据关键用户动态的调整预测模型。在真实的微博网络数据集上的实验表明,模型能够有效地预测消息传播,在预测的同时能够发现影响模型预测效果的关键用户。最后,针对社交网络舆情产生时仅通过删除相关舆情消息并不能有效地对舆情进行控制的问题,提出了面向微博网络用户群体的影响力最大化算法来进行舆论引导。为了降低计算复杂性,通过删除网络无效节点和构建简化的微博网络图来简化网络规模。在简化的微博网络上建立影响力累积贪心算法来选择初始激活集合。通过控制网络中的初始激活集合中的用户发布舆情引导信息,达到控制舆情发展的目的。在真实的微博网络数据集上的实验表明,该方法能够有效地适用于特定目标用户群体的舆情控制。
[Abstract]:In recent years, with the development of social networks, social network has gradually replaced the traditional media, has become the main platform for people to obtain information, dissemination of information. The dissemination of information data in social network is more convenient, in the dissemination of information while providing convenience for rumors, provides a better platform for the dissemination of reactionary information. Public opinion and social reality, social networks can not affected by time, space constraints, social network public opinion makes more difficult to control. When the negative public opinion social network outbreaks, if not promptly and effectively control the negative public opinion, it will become a widespread social event, endanger the security of the whole society. Therefore, research the social network security of public opinion has been focus of academia and industry. Micro-blog social network platform based on the most representative for the study of Like, to study theimpact of social network public opinion dissemination, forecast and control. Because the micro-blog network has a large amount of data and information is fragmented, interactive diversity, rapid transmission of information and other characteristics, spread through the system audit or artificial real-time monitoring method can not effectively restrict the social network public opinion crisis information. Therefore, according to micro-blog the social network public opinion to promote the media, how to realize the automatic transmission effect of micro-blog public opinion measurement, prediction, control and supervision of the spread of negative public opinion information, effectively prevent the outbreak of the crisis and the correct guidance of public opinion micro-blog micro-blog public opinion is an important problem in social security in the network public opinion. This article from the perspective of social security, mainly for the following aspects: first, according to the research data of the existence of a large number of pieces of data of large-scale social networks, and this Some sparse data with high efficient clustering algorithm, combined with a multi entity oriented sparse relational data. In the algorithm, in order to make full use of multi relational data, put forward to build the constraint information embedding method is a robust relation matrix, reduces the sparsity of the matrix, to further improve the accuracy of the algorithm. The block coordinate descent framework under the sparse constraint relation matrix, clustering by non negative matrix factorization algorithm and obtain three different entities indicating matrix. Non negative matrix factorization process, through projection algorithm efficient implementation of fast solution, to ensure that the clustering results show that sparse structure. In artificial and real data experiments, algorithm has significantly improved in the three indicators, especially in the extreme sparsity of data on the effect is more obvious. Secondly, according to the social network public opinion event related news spread influence quantity, Considering the history of the interactive user attributes, user interaction based on the type and frequency of interaction, the user interaction model based micro-blog has spread to the tree. The establishment of user interaction metric matrix to describe the interaction among users in the model, the single message propagation process user interaction diagram to sparse processing form spread to the tree, to depict the news quickly spread process description. Use the interactive tree edge measure matrix measure and influence news spam recognition. Sina micro-blog shows that in real data sets. Experiments on propagation tree model can effectively describe the message propagation, propagation effects and can effectively measure the message. Again, according to social network public opinion event related news propagation prediction problem, considering the influence of public opinion events key user communication, presents a dynamic key based on user Linear prediction model. In order to avoid the influence on structure property prediction message size, message transmission scale portrayed spread scale problem of multiple users, and establish user spread influence function. In order to improve the efficiency of fitting linear model, considering only the key users spread effect to generate linear function prediction model. By using the difference of key users the forecast value and the actual value of the detection of public opinion event message communication, and the prediction model of dynamic adjustment. According to the key users in the real data show that the micro-blog network on a set of experiments, the model can effectively predict the news spread, the forecast also can find the key user models to predict the impact effect. Finally, for only by delete the relevant news public opinion and can not control effectively the problem of public opinion on social network of public opinion, put forward for micro-blog network user groups The influence maximization algorithm to guide public opinion. In order to reduce the computational complexity, to simplify the network size by deleting the invalid network node and simplify the construction of micro-blog network map. To establish a presence in the simplified micro-blog network cumulative greedy algorithm to select the initial activation set. Through the initial activation of the control network in the collection of public opinion to guide users to publish information. To control the development of public opinion. In the real data show that the micro-blog network on a set of experiments, the public opinion control method can effectively apply to the specific target user group.

【学位授予单位】：哈尔滨工程大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：G206;TP393.09

【相似文献】