动态多维社会网络中个性化推荐方法研究
发布时间：2018-10-23 13:41
【摘要】：当前，互联网时代的信息传递已经深刻地改变了人们的信息共享方式，Web已经成为人们获取信息的主要途径。搜索引擎的出现从一定程度上满足了人们信息检索的需求，但它并不能满足不同领域，不同层次用户的需求。个性化推荐技术应信息检索的需求而生，它是个性化服务的一种模式，本质是信息过滤。 个性化推荐系统不仅能在社会经济中发挥巨大的价值，同时也是个非常值得研究的科学问题。目前最为经典的推荐方法是协同过滤推荐，而比较新颖的推荐方法则是基于网络结构的推荐。一般来说，推荐方法都是在单一资源网络中研究用户兴趣，并未过多涉及到多种资源组合成的多维网络，多维网络中的个性化推荐是一个比较新颖的研究角度。 针对上述问题，本文在协同过滤和基于网络结构的推荐思想启发下，在研究社会网络和复杂网络理论的基础上，将多维网络和复杂网络的分析方法引入个性化推荐的研究中来，提出一种动态多维社会网络的个性化推荐方法。首先提出多维交叠网络及其映射网络的定义，构建用户之间多维加权网络模型；在此基础上，引入局域世界演化理论，生成符合本文环境的网络模型演化规则，生成动态多维网络模型；使用识别重叠网络簇的复杂网络聚类算法CPM寻找邻居用户，并最终做出推荐。 本文的主要工作和创新点包括： 1.通过分析社会网络的概念和特点，着重研究多模网络的定义和用户在多模网络中的活动规律，，给出了一个比较清晰的多维交叠网络及其映射网络的数学化定义。尽管多维网络的概念早有学者提出，定义也是多种多样，但目前还没有一个统一的数学化定义。文章通过对多维网络的形成和多维网络转化成一维网络的方法进行研究，归纳总结现有的多维网络形成和降维方法，给出一个并非普适的，但能比较清晰地刻画多维交叠网络及其映射网络形成过程的定义。通过构建用户之间多维加权网络模型的方式来描述参与个性化推荐的用户，改进了原有的只使用兴趣描述文件的用户建模方法。 2.在建立的用户之间的多维加权网络中，分析其具有的复杂网络特性，尤其是局域演化规则。根据经典的局域世界演化理论，以用户之间相似度为节点连接概率因素，改进连接概率公式，提出符合本文用于个性化推荐的多维加权网络的局域世界演化理论模型，并以此生成动态多维网络。动态多维网络模型是进行个性化推荐算法的前提条件，是对用户数据的挖掘和更新模型。 3.使用能识别重叠网络簇的CPM算法进行用户聚类。本文建立的动态多维网络模型具有复杂网络的特征；同时，由于用户兴趣的广泛性和多维交叠网络的特点，在寻找邻居用户群时极有可能发生聚类的重叠。因此，采用能识别重叠聚类簇结构的复杂网络聚类算法寻找邻居用户，符合本文个性化推荐的网络环境。此外，本文还使用了基于用户相似性的最近邻查找方法，并给出了推荐策略。 4.在生成的动态多维网络中进行个性化推荐算法的仿真实验，从不同角度验证了所提算法的有效性。在与常用推荐算法的比较，动态因素的考量以及聚类方法的选择三方面给出了验证结果，并通过不同的评价标准验证了算法的优势性并给出了算法的推荐系统应用模型。
[Abstract]:At present, the information transmission of Internet era has changed people's information sharing mode deeply, and Web has become the main way of people getting information. The appearance of search engine meets the needs of people's information retrieval to a certain extent, but it does not meet the needs of users in different fields and different levels. Personalized recommendation technology should be generated by information retrieval. It is a mode of personalized service. It is the essence of information filtering. Personalized recommendation system can not only play a great value in the social economy, but also a scientific question worth studying at the same time At present, the most classic recommended method is collaborative filtering recommendation, while the newer recommended method is based on the push of the network structure Recommendation. Generally, the recommended method is to study user's interests in a single resource network, not too many multi-dimensional networks combining multiple resources, and the personalized recommendation in multi-dimensional network is a relatively new research angle Aiming at the above problems, based on the research of social network and complex network theory, this paper introduces the method of multi-dimensional network and complex network into the research of personalized recommendation based on the research of social network and complex network theory. In the research, a dynamic multi-dimensional social network personalized push is proposed The method comprises the following steps: firstly, a multidimensional overlapping network and a definition of a mapping network of a multidimensional overlapping network are proposed, a multidimensional weighted network model among users is constructed; based on the method, a local world evolution theory is introduced to generate network model evolution rules which are consistent with the environment, and a dynamic multidimensional network is generated. A complex network clustering algorithm CPM identifying overlapping network clusters is used for finding neighbor users and finally Recommended. The main work of this article and The innovation points include: 1. By analyzing the concept and characteristics of the social network, the definition of the multi-mode network and the rule of activity of the users in the multi-mode network are studied, a relatively clear multi-dimensional overlapping network and its mapping network are given. Although the concept of multi-dimensional network has been put forward by scholars, the definition is also varied, but there is still no system at present This paper studies the formation of multi-dimensional networks and the transformation of multi-dimensional networks into one-dimensional networks, sums up the existing multi-dimensional network forming and dimensionality reduction methods, and gives a non-generalized, but can clearly depict multidimensional overlapping networks and their mapping networks. A user who participates in the personalized recommendation is described by constructing a multi-dimensional weighted network model between users, the original use interest description file is improved, in a multi-dimensional weighted network between established users, analyzing the complex network characteristics it has, in particular to a local evolution rule. According to the classical local world evolution theory, the probability factor is connected with the similarity between the users as the node connection probability factor, the connection probability formula is improved, the local-area world evolution theory model for the multi-dimensional weighted network for the personalized recommendation is proposed and the dynamic multi-dimensional network model is a precondition of carrying out the personalized recommendation algorithm, mining and updating models according to. 3. Use the ability to identify overlapping network clusters The dynamic multi-dimensional network model established in this paper has the characteristics of complex networks, and at the same time, due to the wide range of interests of users and the characteristics of multi-dimensional overlapping networks, Therefore, it is possible to find neighbor users by using complex network clustering algorithms which can identify overlapping cluster structures and conform to the present invention. In addition, a recent neighbor search based on user similarity is also used in this paper In this paper, a simulation experiment of personalized recommendation algorithm is given in the generated dynamic multi-dimensional network. The effectiveness of the proposed algorithm is verified. In comparison with the commonly used recommendation algorithms, the dynamic factors are considered as well as the three aspects of the method of clustering, and the advantages of the algorithm are verified by different evaluation criteria.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.3
本文编号：2289425
[Abstract]:At present, the information transmission of Internet era has changed people's information sharing mode deeply, and Web has become the main way of people getting information. The appearance of search engine meets the needs of people's information retrieval to a certain extent, but it does not meet the needs of users in different fields and different levels. Personalized recommendation technology should be generated by information retrieval. It is a mode of personalized service. It is the essence of information filtering. Personalized recommendation system can not only play a great value in the social economy, but also a scientific question worth studying at the same time At present, the most classic recommended method is collaborative filtering recommendation, while the newer recommended method is based on the push of the network structure Recommendation. Generally, the recommended method is to study user's interests in a single resource network, not too many multi-dimensional networks combining multiple resources, and the personalized recommendation in multi-dimensional network is a relatively new research angle Aiming at the above problems, based on the research of social network and complex network theory, this paper introduces the method of multi-dimensional network and complex network into the research of personalized recommendation based on the research of social network and complex network theory. In the research, a dynamic multi-dimensional social network personalized push is proposed The method comprises the following steps: firstly, a multidimensional overlapping network and a definition of a mapping network of a multidimensional overlapping network are proposed, a multidimensional weighted network model among users is constructed; based on the method, a local world evolution theory is introduced to generate network model evolution rules which are consistent with the environment, and a dynamic multidimensional network is generated. A complex network clustering algorithm CPM identifying overlapping network clusters is used for finding neighbor users and finally Recommended. The main work of this article and The innovation points include: 1. By analyzing the concept and characteristics of the social network, the definition of the multi-mode network and the rule of activity of the users in the multi-mode network are studied, a relatively clear multi-dimensional overlapping network and its mapping network are given. Although the concept of multi-dimensional network has been put forward by scholars, the definition is also varied, but there is still no system at present This paper studies the formation of multi-dimensional networks and the transformation of multi-dimensional networks into one-dimensional networks, sums up the existing multi-dimensional network forming and dimensionality reduction methods, and gives a non-generalized, but can clearly depict multidimensional overlapping networks and their mapping networks. A user who participates in the personalized recommendation is described by constructing a multi-dimensional weighted network model between users, the original use interest description file is improved, in a multi-dimensional weighted network between established users, analyzing the complex network characteristics it has, in particular to a local evolution rule. According to the classical local world evolution theory, the probability factor is connected with the similarity between the users as the node connection probability factor, the connection probability formula is improved, the local-area world evolution theory model for the multi-dimensional weighted network for the personalized recommendation is proposed and the dynamic multi-dimensional network model is a precondition of carrying out the personalized recommendation algorithm, mining and updating models according to. 3. Use the ability to identify overlapping network clusters The dynamic multi-dimensional network model established in this paper has the characteristics of complex networks, and at the same time, due to the wide range of interests of users and the characteristics of multi-dimensional overlapping networks, Therefore, it is possible to find neighbor users by using complex network clustering algorithms which can identify overlapping cluster structures and conform to the present invention. In addition, a recent neighbor search based on user similarity is also used in this paper In this paper, a simulation experiment of personalized recommendation algorithm is given in the generated dynamic multi-dimensional network. The effectiveness of the proposed algorithm is verified. In comparison with the commonly used recommendation algorithms, the dynamic factors are considered as well as the three aspects of the method of clustering, and the advantages of the algorithm are verified by different evaluation criteria.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.3
【引证文献】
相关期刊论文 前1条
1 郭平;刘波;沈岳;;农业云大数据自组织推送关键技术综述[J];软件;2013年03期
本文编号：2289425
本文链接：https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/2289425.html
