社会化媒体内容关注度分析与建模方法研究
发布时间：2019-06-05 17:46
【摘要】：社会化媒体近年来得到极大发展，已经在整个互联网中占据主流地位。根据世界著名流量统计网站Alexa的数据，全球访问量排名前十的网站中，有五个是社会化媒体网站。社会化媒体的空前发展和应用，孕育了大量新的研究领域，比如催生了新的信息技术研究，促进了针对人类社会行为规律的理论研究。2009年Science杂志发表了题为《计算社会学》的文章，标志着计算科学和社会科学的交叉领域已成为国际前沿研究热点，而社会关注度是其中最为重要的研究领域之一。社会关注度分布及动态增长特性的研究不仅能够加深对人类宏观行为规律的理解，而且对于理解和提升诸如预取缓存、P2P网络、搜索引擎和推荐系统的性能具有重要的理论价值。本文在社会关注度分布特征分析、社会关注度传播过程特性、基于社会关注度分布特性的预取缓存技术以及提高社会关注度方法等问题上进行了深入的研究。 首先，分析了多来源社会关注度分布的若干特征以及各来源对社会关注度分布的影响。社会化媒体内容规模巨大，并且具有高度动态性和高度分散性的特点，可能使得传统的分布模型和预测方法失效。本文从全局和局部两个层面同时对多来源社会关注度整体分布特征进行了分析，发现了全局和局部社会关注度分布的差异。在此基础上，深入分析了不同来源对社会关注度分布的影响，，结果表明搜索引擎和推荐系统是社会关注度的两大主要来源，并且搜索引擎倾向于加剧“马太效应”，而推荐系统则有助于减轻“马太效应”。该研究成果有助于回答学术界所广泛关心的搜索引擎和推荐系统如何影响被观看媒体内容多样性的问题。 其次，提出了基于用户行为模型聚类(Clustered User Behavior Model, CUBM)的媒体对象预取缓存方法。本文借助PlanetLab平台测量和分析了社会化多媒体网站在传送大尺寸多媒体对象时出现频繁中断的问题，论述了采用预取缓存技术的必要性。在此基础上，提出一种基于用户行为模型聚类(CUBM)的媒体对象预取缓存方法。该方法将行为模式类似的用户归类并分别建立Markov链，克服了传统方法未能体现用户差异以及在局部代理部署时覆盖率不高的缺点，并且抓住了活跃用户比不活跃用户倾向于观看更多内容的事实，从而提高了预取的准确率和命中率。 再次，提出了基于随机游走的社会关注度传播模型(Random Walk based PopularityPropagation Model, RWPPM)。为了深入理解媒体对象如何通过媒体对象关系网影响对方的社会关注度，本文提出了一个基于随机游走的社会关注度传播模型。随后分析了模型的收敛条件，论述了模型的功能并验证了模型的正确性。在此基础上，运用RWPPM模型对YouTube视频网络中视频间社会关注度的相互影响力及其特征进行了分析。 最后，提出了一种基于KTK (Keywords-Topics-Keywords)关键词推荐的社会关注度提高方法。分析了媒体对象标识文本关键词在搜索引擎检索和推荐系统推荐媒体对象中的重要性。进而研究了媒体对象关系网的簇结构以及各簇主要关键词代表话题的能力。在此基础上，提出一种遵循“关键词—主题—关键词”思路，兼顾相关度和社会关注度的KTK关键词推荐算法。最后，实验结果表明所推荐关键词能够大幅提高媒体对象的社会关注度。
[Abstract]:The social media has been greatly developed in recent years and has been in the mainstream in the whole Internet. According to Alexa's data from the world's well-known traffic statistics website, there are five social media sites in the top 10 of the world's top-ranked websites. The unprecedented development and application of the social media has given birth to a large number of new research fields, such as the emergence of new information technology research and a theoretical study on the law of human society's behavior. In 2009, the Science magazine published an article entitled "Computing sociology>," It is one of the most important research fields in which the cross-cutting area of computing science and social science has become a hot spot in the international front, and the degree of social attention is one of the most important research fields. The study of the distribution of social attention and the characteristics of dynamic growth not only can deepen the understanding of the law of human macro-behavior, but also has important theoretical value to understand and improve the performance of such as pre-fetch cache, P2P network, search engine and recommendation system. In this paper, the characteristics of the distribution of the social attention, the characteristics of the communication process of the social attention, the prefetch buffer technology based on the distribution characteristics of the social attention and the methods of improving the social attention are deeply studied in this paper. First of all, the characteristics of the distribution of the multi-source social attention and the shadow of the distribution of the social attention of each source are analyzed. In response, the content of the social media is large, and it has the characteristics of high dynamic and high dispersivity, which can make the traditional distribution model and the prediction method to be lost. In this paper, the overall distribution characteristics of the multi-source social attention are analyzed from both the global and the local aspects, and the difference between the global and local social attention distribution is found. On this basis, the influence of different sources on the distribution of social attention is analyzed. The results show that the search engine and the recommendation system are the two main sources of the social attention, and the search engine tends to aggravate the "Matthew's effect", and the recommendation system will help to reduce the "big>" Matthew's effect ". The research results help to answer the question of how the search engine and the recommended system of the academic community have an impact on the diversity of the content of the media to be viewed Secondly, a media object pre-fetch delay based on the user behavior model (CUBM) is proposed. In this paper, using the PlanetLab platform to measure and analyze the problem of frequent interruption of the social multimedia web site in the transmission of large-size multimedia objects, this paper discusses the use of the prefetch buffer technology. In this paper, a kind of media object pre-fetching buffer based on user behavior model clustering (CUBM) is put forward. The method comprises the following steps of: classifying and establishing a Markov chain by a user similar to the behavior mode, the fact that the accuracy of pre-fetching is improved and In this paper, a random walk-based social attention propagation model (R) is presented. WPPM (WPPM). In order to understand how the media objects influence the social attention of each other through the media object relationship network, this paper puts forward a social concern based on random walk In this paper, the convergence condition of the model is analyzed, the function of the model is discussed, and the model is verified. On the basis of this, the mutual influence and characteristics of the social attention of the video in the YouTube video network by the RWPPM model Finally, a society based on KTK (Keyworld-Topics-Keywords) keyword is proposed. The method for improving the attention degree is analyzed, and the media object identification text keyword is analyzed in a search engine search and recommendation system recommendation medium The importance of the object is also studied. The cluster structure of the media object and the key words of each cluster are also studied. On the basis of this, this paper puts forward a kind of KTK which follows the "Key words and key words" of thinking, takes into account the degree of correlation and the degree of social attention. Finally, the experimental results show that the recommended keywords can greatly improve the media pair.
【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】：2012
【分类号】：TP393.0
本文编号：2493710
[Abstract]:The social media has been greatly developed in recent years and has been in the mainstream in the whole Internet. According to Alexa's data from the world's well-known traffic statistics website, there are five social media sites in the top 10 of the world's top-ranked websites. The unprecedented development and application of the social media has given birth to a large number of new research fields, such as the emergence of new information technology research and a theoretical study on the law of human society's behavior. In 2009, the Science magazine published an article entitled "Computing sociology>," It is one of the most important research fields in which the cross-cutting area of computing science and social science has become a hot spot in the international front, and the degree of social attention is one of the most important research fields. The study of the distribution of social attention and the characteristics of dynamic growth not only can deepen the understanding of the law of human macro-behavior, but also has important theoretical value to understand and improve the performance of such as pre-fetch cache, P2P network, search engine and recommendation system. In this paper, the characteristics of the distribution of the social attention, the characteristics of the communication process of the social attention, the prefetch buffer technology based on the distribution characteristics of the social attention and the methods of improving the social attention are deeply studied in this paper. First of all, the characteristics of the distribution of the multi-source social attention and the shadow of the distribution of the social attention of each source are analyzed. In response, the content of the social media is large, and it has the characteristics of high dynamic and high dispersivity, which can make the traditional distribution model and the prediction method to be lost. In this paper, the overall distribution characteristics of the multi-source social attention are analyzed from both the global and the local aspects, and the difference between the global and local social attention distribution is found. On this basis, the influence of different sources on the distribution of social attention is analyzed. The results show that the search engine and the recommendation system are the two main sources of the social attention, and the search engine tends to aggravate the "Matthew's effect", and the recommendation system will help to reduce the "big>" Matthew's effect ". The research results help to answer the question of how the search engine and the recommended system of the academic community have an impact on the diversity of the content of the media to be viewed Secondly, a media object pre-fetch delay based on the user behavior model (CUBM) is proposed. In this paper, using the PlanetLab platform to measure and analyze the problem of frequent interruption of the social multimedia web site in the transmission of large-size multimedia objects, this paper discusses the use of the prefetch buffer technology. In this paper, a kind of media object pre-fetching buffer based on user behavior model clustering (CUBM) is put forward. The method comprises the following steps of: classifying and establishing a Markov chain by a user similar to the behavior mode, the fact that the accuracy of pre-fetching is improved and In this paper, a random walk-based social attention propagation model (R) is presented. WPPM (WPPM). In order to understand how the media objects influence the social attention of each other through the media object relationship network, this paper puts forward a social concern based on random walk In this paper, the convergence condition of the model is analyzed, the function of the model is discussed, and the model is verified. On the basis of this, the mutual influence and characteristics of the social attention of the video in the YouTube video network by the RWPPM model Finally, a society based on KTK (Keyworld-Topics-Keywords) keyword is proposed. The method for improving the attention degree is analyzed, and the media object identification text keyword is analyzed in a search engine search and recommendation system recommendation medium The importance of the object is also studied. The cluster structure of the media object and the key words of each cluster are also studied. On the basis of this, this paper puts forward a kind of KTK which follows the "Key words and key words" of thinking, takes into account the degree of correlation and the degree of social attention. Finally, the experimental results show that the recommended keywords can greatly improve the media pair.
【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】：2012
【分类号】：TP393.0
【引证文献】
相关硕士学位论文 前2条
1 罗跃红;“高度关注”下乡村艾滋病人社会行动策略研究[D];华中师范大学;2013年
2 罗燕妮;社会化媒体环境中的口碑传播研究[D];华南理工大学;2013年
本文编号：2493710
本文链接：https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/2493710.html
