【作者】 武慧娟；

【导师】 徐宝祥；

【作者基本信息】 吉林大学 ， 情报学， 2014， 博士

【摘要】 在Web3.0时代，人们不再满足由机器挖掘给用户的各种信息，而是要结合自身的偏好享用个性化的信息服务，无论是产品还是服务，都将为每一用户量身打造，信息世界将变得越来越聪明和智能，似乎比用户还了解他想要的是什么，这就是个性化的信息服务，即Web3.0的内涵。所以，如何在浩如烟海的信息中寻求个性化的信息，就是目前学术界需要解决的热点研究问题。同时，在另一方面，随着Internet网络的产生和使用，用户在社会化标注系统(Web页面)上产生了大量的代表自己个性化信息的数据，那么如何充分利用用户产生的数据来解决目前的个性化信息服务问题，就成了一个情报学术界研究的热点学术问题。社会化标注系统主要是以“用户-资源-标签”三元关系为研究对象的典型复杂动态网络，用户可以根据个人的需要自由选择词汇对所喜爱的资源进行标注，每添加一个词汇被称为对资源添加一个标签，用户、资源和标签组成了社会化标注系统的三个基本元素。从这三个最基本的组成元素开始，从中提炼用户的个性化信息，进而形成个性化信息推荐，成为论文的研究起点。本文针对社会化标注系统中的个性化信息推荐模型问题展开研究，在综述“社会化标注系统”、“个性化信息推荐”的基础上，对国内外“社会化标注系统中的个性化信息推荐模型”研究现状进行分析，通过自组织理论、社会网络分析理论、系统动力学等理论与方法围绕社会化标注系统的演化过程、用户关系网络、个性化信息推荐模型构建等核心问题进行详细分析。本文研究的内容主要分为五大部分：内容一：研究目前国内外社会化标注系统、个性化信息推荐两方面的研究进展，分析研究的热点与前沿问题，掌握在具体的研究中其所面临的不足，然后从不足处展开，形成本文所要研究的起点，即利用社会网络分析研究用户关系网络，作为分析个性化信息推荐的逻辑起点。内容二：研究社会化标注系统的演化形式及耗散结构。对社会化标注系统中的用户、资源、标签及三者之间的关系进行分析，利用自组织理论研究社会化标注系统的演化机理，具体使用超循环和耗散结构分别分析系统中的用户、标签、资源各自的演化形式及系统的多层级耗散结构。内容三：分析社会化标注系统中的用户关系网络。主要分析了用户关系网络的结构特征，如网络密度、核心-边缘结构、中心性等，同时，从用户关系网络中挖掘凝聚子群的各种信息偏好，从两个角度：群内和群际对信息进行分析，发现个性化的信息。内容四：构建社会化标注系统中的个性化信息推荐模型。从静态、动态两个角度分别构建社会化标注系统中的个性化信息推荐模型；然后利用系统动力学绘制个性化信息推荐的系统模型，主要由系统的因果关系和系统流图两个步骤构成。内容五：对个性化信息推荐进行实证研究，首先从豆瓣网中抓取数据，然后对其进行自组织演化、用户关系网络结构分析、个性化信息推荐模型构建等。本文的创新点体现在如下两个方面：第一，研究视角的创新。本文从社会网络的角度研究在用户标注行为中产生的用户关系网络，从中提取个性化信息，并从群内和群际两个角度进行个性化信息推荐分析。当前的相关研究主要局限于社会化标注系统的网络结构分析，少部分研究涉及到了凝聚子群，群内和群际信息推荐的研究成果几乎没有。因此，本文为个性化信息推荐问题的研究引入了全新的分析视角。第二，研究内容的创新。本文采用“移植”借鉴的方法，通过科学“移植”信息科学领域的经典理论，对社会化标注系统中的个性化信息推荐模型进行深入研究，创新认识，具体体现在：根据自组织理论提出了社会化标注系统的超循环演化形式和多层级耗散结构，深化了系统演化的过程，为模型建立打下了深厚的理论基础；根据社会网络分析理论提出了基于群内的个性化信息推荐和基于群外的个性化信息推荐两种推荐方法，深化了方法论思想；构建了个性化信息推荐模型的静态和动态模型，并运用系统动力学对个性化信息推荐模型进行动力学影响因素分析，提高了模型化的程度。本文的研究结论具体如下：第一，基于自组织理论的社会化标注系统的演化形式遵循超循环理论，同时具有层级耗散结构特征。社会化标注系统具有自组织系统的开放性、远离平衡态、非线性相关性、随机涨落等特征；然后运用超循环理论分别对用户集、资源集、标签集及社会化标注系统的自组织演化机理进行了探讨，并构建了社会化标注系统的自组织演化模型；最后根据耗散结构理论，从多层级的角度，对社会化标注系统的序化形成过程进行研究，提出了系统序化的平衡极点。第二，从社会网络分析的角度，可以发现社会化标注系统的用户关系网络结构和凝聚子群。主要分析了用于关系网络的网络密度、核心-边缘结构、中心性等，还分析了用户关系网络的凝聚子群，利用块模型的“结构对等性”对行动者进行聚类，利用K-核分析对块模型进行补充，利用结构洞方法发现存在结构洞的用户，即基于群际的用户。第三，社会化标注系统中的个性化信息推荐模型可以分为静态模型和动态模型来构建。首先从系统学的角度研究构成个性化信息推荐系统的六大要素；然后在遵循构建概念的原则下，构建了个性化信息推荐的静态模型和动态模型；最后利用系统动力学对个性化信息推荐的模型进行影响因素分析，即当网络的密度较小时，基于群内的个性化信息推荐与基于群际的个性化信息推荐受网络密度的影响较大；当网络的密度较大时，基于群内的个性化信息推荐受核心人物的影响较大，基于群际的个性化信息推荐受图的中心势值影响较大。更多还原

【Abstract】 In the Web3.0era, people no longer satisfy various kinds of informationby machine mining, but to combine its own preference for personalized informationservices. Products and services will be tailored for each user, the informationworld will become more and more intelligent and smart, it seems that theinformation world knows more than the users to understand what he wants, andthis is the personalized information services, namely the connotation of Web3.0. So, how to seek personalized information in the voluminous information ispresently the research hot problems need to be solved. At the same time, on theother hand, with the use of the Internet, users produced a large number of datathat they are on their own individual information data in the social taggingsystem (Web page), so how to make full use of the data generated by the userto solve the problem of the individualized information service, became a hotacademic intelligence of the academic research.Social tagging system is a typical complex dynamic network that it mainlybased on "user-resources-label" ternary relationship, users can tag theresources of favorite according to the needs of individual, freely choice ofthe resources, each adding a word is called to add a label for resources. Theusers, resources and tags become the three basic elements of social taggingsystem. The paper start the study from the basic composition of the three elements,and extract for the user’s personalized information, then personalizedinformation recommendation.This paper launches the model research of personalized informationrecommendation in social tagging system, analyzing the present situation andproblems of research on the basis of the domestic and foreign personalizedinformation recommendation model,based the review of the "social taggingsystem" and "personalized information recommendation". The paper analyzes the processes of the social tagging system, user relationship network structure,personalized information recommendation model building through theself-organization theory, social network analysis theory, and system dynamicstheory.This paper studies the main content is divided into five:Content one: The paper studies the progress of social tagging systems andpersonalized information recommendation at home and abroad, and analyzes thehot and frontier issues, masters the shortage in the study, and then formed thestarting point of the study in this paper, beginning from the insufficient place,using the social network analysis to study the user network. This is the logicalstarting point of analysis of personalized information recommendation.Content two: The paper studies the evolution forms and dissipative structureof social tagging system.It analyzes the there relations in the social tagging system, and studiesthe evolution of the social tagging system mechanism, using theself-organization theory.they are the super cycle and the dissipative structure,and this paper analyzes the evolution forms in system user, tag, resources andthe system’s multi-level dissipative structure.Content three: The paper studies the use network in the social taggingsystem.It Mainly analyzes the structure characteristics of the users network, suchas network density, the core-edge, centricity, etc., at the same time,subgrouping the preference from the user network, to analyze the informationfrom two perspectives: group and intergroup, and find per sonalized information.Content four: The paper builds a model of personalized informationrecommendation in the social tagging system.It sets up the model of personalized information recommendation in the socialtagging system from two aspects: static and dynamic respectively, and then drawsthe personalized information recommendation system using the system dynamics model, mainly consists of two graphics: the system of cause and effectrelationship, the system flow chart of two steps.Content five: The paper makes the empirical research of personalizedinformation recommendation. Firstly it fetches all data from Dou Ban, thenanalyzes the system self-organization evolution, the network structure of usernetwork, and builds the model of personalized information recommendation.This paper’s innovations embodied in two aspects:Firstly,the innovation of research angle.This paper studies the user network produced the act of user annotationsfrom the perspective of social network, extracting the personalized information,and analysising the personalized information recommendation from two aspects:group and intergroup. The current related research mainly is limited in thesocial tagging system’s network structure analysis, a few studies involvingcondensing subgroup,and hardly studies in the same group and intergroupinformation recommended. Therefore, this paper introduced the new perspectivein the research of personalized information recommendation problem.Secondly,the innovation of research content.This paper studies the model of personalized information recommendation inthe social tagging system in-depth, adopting the method of "transplant", throughscientific "transplant" the classic theory in the field of information science,specific include:according to the theory of self-organization, it analyzes theevolution of social tagging system in detail，to deepen the process of systemevolution and lay a profound theoretical foundation of establishing a model；according to the theory of social network analysis, it presents tworecommendations based on the group and group outside of the personalizedinformation recommendation，to deepen the thoughts of methodology；it buildsthe static and dynamic model of the personalized information recommendation,and then analyzes the whole dynamic simulation of the model using systemdynamics,to increase the degree of the model. This paper’s research conclusion is as follows:Firstly, the evolution form of social tagging system follows the hypercycletheory based on the theory of self-organization, at the same time has thecharacteristics of dissipative structure hierarchy.The social tagging systemhas the self-organizing system of openness, far from equilibrium, nonlinearcorrelation and random fluctuation characteristics; Then discusses theself-organization evolution mechanism of the social tagging system usinghypercycle theory respectively on the user set, resource set, the label set andthe social tagging system; According to the dissipative structure theory, fromthe perspective of a multi-level, studying the ordering of social tagging system,putting forward the system order balance pole.Secondly, we can find the social tagging system user relationship networkstructure and condensing subgroup from the perspective of social networkanalysis. It mainly analyzes the network of network density, the core-edgestructure, centricity, and condensing subgroup and so on, using the block modelstructure of "equivalence" to clustering of actors, using the K-nuclear analysisto supplement the block model, using the method of structural holes to find thehole structure of users, based on intergroup users.Thirdly, the personalized information recommendation in the social taggingsystem model can be divided into static model and dynamic model. First of all,it analyzes the six key elements in personalized information recommendationsystem from the angle of systematics; then under the follow the principle ofbuilding concept, it builds a static model and dynamic model of personalizedinformation recommendation; Finaly it builds the system dynamics simulationmodel of personalized information recommendation system analysis. When thenetwork density is small, the personalized information recommendation based onthe group and the personalized information recommendation based on intergroupis strongly influenced by the network density; When the network density is larger,the personalized information recommendation based on the group is greatly influenced by the central figure, the personalized information recommendationbased on intergroup is greatly influenced by figure at the center of the potentialvalue.更多还原