基于网络用户行为的搜索排行榜研究

发布时间：2018-04-06 00:32

本文选题：搜索排行榜　切入点：MSC　出处：《北京交通大学》2013年博士论文

【摘要】：互联网已经成为人类社会最重要的信息基础设施。“搜索引擎”与“网络音乐”无论是从发展速度还是从用户规模上均处于网络应用的前列。音乐搜索排行榜跨越这两大应用,其用户规模大,分析和研究其用户行为对推动互联网应用的发展有重要意义。本文以音乐搜索排行榜作为研究对象,从搜索排行榜的特征、影响因素、用户的反作用以及社会学结构等方面研究了搜索排行榜的相关特性。论文工作基于真实网络环境,在网络测量的基础上,从时间序列、分类特征以及榜单的变化等方面针对实际数据进行多角度分析,以理论推导和仿真实验为手段研究搜索排行榜与网络用户之间的相互作用；根据排行榜的时间特性,提出了改进的团结构查找算法——权重标签繁殖算法,进而通过社会关系网络分析用户行为的拓扑及动态特性。针对搜索排行榜的相关研究,本文的主要工作与创新点如下： 1.建立了一种描述搜索排行榜影响力的数学模型。为了研究搜索排行榜对网络用户行为的影响作用,创新性地提出影响因子和影响时间这两个参数评价影响力,并以这两个参数为基础,针对搜索排行榜的影响力进行数学建模。经过真实数据和仿真实验两种方法的验证,证实了模型的有效性。由该模型可得出两个结论：一是初始阶段排行榜体现了用户的意愿,二是在一定时间范围内排行榜与用户意愿有差异,而该时间的长短受到两个因素的影响,它们分别是排行榜的影响力和排行榜与用户意愿之间的差异。 2.建立了一种描述排行榜“刷榜”现象的数学模型。网络用户中的一类特殊用户,他们在一段时间内对某个关键词进行集中搜索,这一类特殊行为定义为“刷榜”行为,这种现象称为“刷榜”现象。为了研究网络用户行为对搜索排行榜的影响作用,从用户的特殊搜索行为入手,针对“刷榜”现象的影响力进行建模,得出其影响力的数学描述,并通过仿真验证了模型的有效性。对数学模型的分析表明,“刷榜”的影响力和排行榜影响力之间存在一种相互制约、相互促进的关系,在两种因素的作用下,排行榜的变化具有很强的随机性。 3.提出了一种团结构查找算法——改进的权重标签繁殖算法(WLPA)。通过对搜索排行榜社会网络基本特征的分析,发现现有算法无法进行合理的团结构划分,在标签繁殖算法的基础上,创新性地引入了边权重的概念,提出基于权重的标签繁殖算法。改进的权重标签繁殖算法在原有标签繁殖算法的基础上,将边的权重作为衡量节点间关系的依据,据此进行团结构的划分。改进算法可以适应排行榜网络边权重大于一的特点,并对这类社会学网络具有普适性。和原算法相比,改进算法具有线性复杂度,运行高效快速。 4.发现了MSC网络的团结构特征。为了研究网络用户的团结构特征,采用网络团结构查找分析方法,以音乐搜索排行榜的用户行为数据为基础,对用户的社会关系网络进行分析。分析结果表明：(1)在网络中存在核心节点,(2)网络结构具有明显的时间特性。由于关键词的搜索点击行为带有明显的倾向性,能够反映用户的个人喜好,MSC网络的团结构特征就是其用户行为的特征。MSC的用户也具有核心用户与时间特性的特点。MSC团结构的特点是进行用户分类和用户推荐研究的理论依据。
[Abstract]:The Internet has become the most important information infrastructure in human society. Whether it is from the development speed and scale from the user are at the forefront of the application of network search engine "and" network music ". Music search charts across these two applications, the user scale, the analysis and research of user behavior has important significance for the development of to promote the Internet application.
The article is based on the music search charts as the research object, the influence factors from the characteristics of the search rankings, the reaction of users as well as the social structure and other aspects of the relevant characteristics of the search rankings. Based on the real network environment, based on network measurement, from the time series, the classification features and changes in the list etc. multi angle analysis on the actual data, the theoretical derivation and simulation experiment method to study the interaction between the search and ranking of network users; according to the time characteristic of the list, put forward the improved algorithm to find cluster weight label propagation algorithm, topology and analysis of user behavior through social networks and dynamic characteristics.
The main work and innovation of this paper are as follows:
1. to establish a mathematical model to describe the search rankings influence. In order to study the search rankings effect of network user behavior, put forward the evaluation factors and effects of these two parameters influence innovation, and taking the two parameters as the basis, mathematical modeling for the search rankings after impact. Validation of real data and simulation results of the two methods, confirmed the validity of the model. The model can draw two conclusions: one is the initial stage of the list reflects the user's intention, the two is in a certain range of time and ranking of users are different, and the length of time is influenced by two factors and they are the difference between the rankings and ranking of the influence and the wishes of the user.
2. to establish a mathematical model to describe the list "brush list" phenomenon. A special class of users in Internet users, they are on a keyword in a period of time to focus the search, this kind of special behavior is defined as "brush list" behavior, this phenomenon is called "brush to list" phenomenon. To study the effect of network user behavior impact on search rankings, starting from the user's search behavior, according to the "brush list" phenomenon of the influence that the influence of modeling, mathematical description, and the validity of the model is verified by simulation analysis. The mathematical model shows that there is a mutual restriction between the brush list "the list of influence and influence, promote each other, in the two kinds of factors, has strong randomicity change list.
3. proposed a cluster search algorithm - improved weight label propagation algorithm (WLPA). Through the analysis of the basic characteristics of social network search rankings, found that the existing algorithm to cluster a reasonable division, based on label propagation algorithm, innovatively introduced the concept of edge weight, weight forward propagation based on the label algorithm. Basic weight improved breeding algorithm label breed in the original tag on the side as a measure of weight between nodes on the basis of the relationship, according to division of group structure. The improved algorithm can adapt to network list edge weight greater than one, and it is a universal learning network of this kind of society. Compared with the original algorithm, the improved algorithm has linear complexity, fast and efficient operation.
4. it is found that the cluster structure characteristics of the MSC network. In order to research on the structural characteristics of group of network users, the network cluster analysis method to search, music search user behavior data list based on the analysis of social network users. Analysis results show that: (1) the existence of core nodes in the network, (2) network structure has obvious time characteristics. Because the search keyword click behavior with a clear bias, which can reflect the user's personal preferences, the characteristics of.MSC structure characteristics of the MSC network is the user behavior of users with the characteristics of.MSC group structure and time characteristics of core users is the theoretical basis for user classification and user recommendation research.

【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP391.3

【相似文献】