基于关键词关联语义链网络的Web事件演化研究

发布时间：2018-01-12 20:38

本文关键词：基于关键词关联语义链网络的Web事件演化研究　出处：《上海大学》2016年博士论文　论文类型：学位论文

【摘要】：在现代社会中,各种热点和突发事件频发,尤其是在中国社会的高速信息化以及国内外各种复杂因素的影响下,社会中的热点和突发事件更是此起彼伏。Web的到来,一方面,加深了热点和突发事件对现实社会的影响力。Web打破了传统信息传播的时间和地域界限,更多的人可以通过Web近乎实时的了解世界上任何地方发生的事件,从而一个事件的影响力较之从前会被Web极大的加深。甚至很多事件的元素都发起自Web并在其中得到广泛报道,进而影响到现实社会中相应的人或组织。由于其的强大的影响力,因此对Web事件的研究势在必行。另一方面,Web也为热点和突发事件的研究提供了可能。通过Web,更多的人可以更加容易的关注、了解并参与到热点和突发事件中来,他们在Web中的“足迹”(即,Web上大量的网页信息)为Web事件的研究提供了丰富的原始资料,其体量与覆盖面都是采用调查与访问得到的数据所不能比拟的。因此,本文欲借助于Web上大量的网页信息,对Web事件进行研究。为了捕捉和整合散落在各个网页中的Web事件的语义信息,本文首先构建了关键词关联语义链网络作为Web事件整体的、统一的语义表示模型,该模型为本文的研究提供了一个重要的理论与计算基础。进而,本文从“静态”和“动态”两个方面对Web事件的演化进行研究。一方面,在关键词关联语义链网络的基础上,本文研究了Web事件某一时刻的演化状态,即子话题的挖掘以及不同网站对这些子话题的偏好挖掘。具体包括:1)基于关键词关联语义链网络,提出了文本图语义表示方法,以提高对网页的语义捕捉能力。进而,提出了面向文本图表示的子话题挖掘模型,以挖掘到Web事件的不同子话题,从而把握Web事件当前时刻的演化状态;2)基于网站-网页-关键词三层关联语义链网络,挖掘网站对Web事件子话题的偏好。为此,本文提出了两种不同的思路:一种是显式的利用三层网络,采用社区划分的思想来挖掘网站偏好;另一种是隐式的利用三层网络,采用概率图模型的方法来挖掘网站偏好。另一方面,基于关键词关联语义链网络,研究了Web事件的演化动力。具体包括:1)基于关键词关联语义链网络,通过关键词权重分布的熵定义并度量了Web事件的不确定性,其中关键词的权重的计算提出了三种不同的策略:第一种仅考虑关键词的统计特性;第二种考虑关键词的统计特性与局部网络结构特性;第三种考虑关键词的统计特性与全局网络结构特性。然后,基于关键词的不确定性的层次性,构建Web事件的语义金字塔模型。最后,将构建的Web事件的语义金字塔模型应用到不同功能的网页推荐中。2)基于关键词关联语义链网络的不确定性以及完美无标度分布,定义并量化了Web事件的不确定性空间,包括两个极端状态:最确定状态(事件关键词关联语义链网络的完美无标度分布)和最不确定状态(事件关键词关联语义链网络的熵)。进而,定义并度量了Web事件演化内部力,即Web事件当前所处的不确定性空间中的位置。3)基于社会模式理论与人类关联记忆理论,对Web事件所在的社会环境进行建模,从而解释并量化了社会环境对Web事件演化的影响。然后,将Web事件与社会环境的交互模拟为两个过程:一个激活过程和一个反馈过程。最后,基于对交互过程的模拟,提出了社会环境影响力的度量方法,即Web事件演化外部力的计算。实验证实两种定义的演化动力都与真实的Web事件演化有着较强的相关性。本文的研究不仅可以用于提供新的面向事件的Web服务,以辅助用户更加容易的了解事件的演化信息;同时也可以辅助国家应急管理部门对Web事件进行及时、准确的研判,以减小Web事件引起的恐慌以及造成的生命财产损失。
[Abstract]:In modern society, a variety of hot and frequent emergencies, especially in the information society and the impact of high-speed China abroad all kinds of complex factors, and the focus of social emergencies is as one falls, another rises the arrival of.Web, on the one hand, deepened the influence of.Web on social hot and unexpected events breaking the time and the traditional boundaries of information dissemination, more and more people can use Web near real-time understanding of any place in the world events, and an event influence than before Web will be greatly deepened. Even many event elements are initiated from Web and widely reported in them, and then influence the people or the organization in the realistic society. Because of its powerful influence, so the research on Web events is imperative. On the other hand, Web is also a research hotspot and unexpected events can provide. Web, more people can more easily understand and participate in the focus of attention, and in the event of an emergency, their footprint in Web "" (i.e., Web on the number of pages of information) provides rich material for the study of Web events, and the amount of coverage is obtained by the investigation and visit the data can not be compared. Therefore, this paper wants to use a lot of information on the Web page, to study the events of Web. In order to capture the semantic information integration and scattered in every page in the Web event, this paper constructs the semantic link network associated keywords as Web events overall, unified semantic representation model. This model provides an important theoretical and computational basis for this research. Then, this article from the "static" and "dynamic" two aspects of Web event evolution are studied. On the one hand, the keywords associated semantic chain network Based on the network, this thesis studies the evolution of Web events at a certain time, the mining sub topic and different site preferences for these sub topic mining. Including: 1) keywords associated semantic link network based on text and image semantic representation method is proposed to improve the semantic web capture ability. Then, the text oriented graph sub topic mining model, to dig into the different sub Web event topic, so as to grasp the evolution of Web events at the moment; 2) - three - page website keywords related semantic link network based on the excavation site of the Web event sub topic preference. Therefore, this paper puts forward two different ideas: one is the three layer network utilization, using community partition of the mining site preference; the other is a three layer network using implicit method, using probabilistic graphical models to other mining site preferences. On the one hand, the semantic link network based on related keywords, study the dynamic evolution of Web events. Including: 1) keywords associated semantic link network based on the entropy weight of keywords distribution and measure Web uncertainties, the calculation of the weight of keywords and put forward three different strategies: first consider only keywords statistical characteristics; second keywords and taken account of the statistical characteristics of local network structure; third keywords and taken account of the statistical characteristics of global network structure. Then the level of uncertainty based on keywords, semantic model construction Pyramid Web event. Finally, the application of semantic model of Pyramid built Web into different events the function of web page recommendation.2) scale-free distribution based on the keywords associated semantic link network uncertainty and perfect, define and quantify the uncertainty of the Web event The qualitative space, including two extremes: the most determined state (perfect event keyword associated semantic link network scale-free distribution) and the uncertainty (event keywords associated semantic link network entropy). Furthermore, the definition and measure the Web event evolution of internal force, the uncertainty of the event space Web the position of.3) Social Theory and human associative memory model based on the theory of modeling of the Web event where the social environment, so as to explain and quantify the effects of social environment on the evolution of Web events. Then, the interaction model of Web events and social environment into two processes: an activation process and a feedback process. Finally, simulation of the interaction process based on the proposed method to measure the influence of social environment, computing Web event evolution of external force. Experiments prove that the two definitions are dynamic evolution and real Web events There is a strong correlation. This study not only can be used to provide event for the new Web service, the evolution of the information easier to assist the user to understand events; at the same time also can assist the national emergency management departments of Web events are timely, accurate judgments, to reduce the Web incident caused panic and the life and property caused by the the loss.

【学位授予单位】：上海大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.1;TP393.092

【相似文献】