网络论坛文章数序列的自相似性建模与预测

【作者】 范文嵩；

【导师】 张世永；

【作者基本信息】 复旦大学 ， 计算机应用技术， 2010， 硕士

【摘要】 随着Web2.0概念的提出,高速变化的互联网正得到人们越来越多的关注。网络论坛、博客等沟通交流形式已经成为互联网应用的热点。一方面,在网络论坛研究中,话题热点分析是一个主要的研究方向；另一方面,由于各种商业活动决策(如产品市场调查或广告投放)的需要,宏观上了解论坛或博客用户参与度的变化正在成为一个新的研究热点,对论坛中文章数随时间变化规律的研究最近也得到了人们的关注。因此,网络论坛特性的精确测量和刻画,对分析、理解和仿真网络论坛的动态变化,对指导网络论坛控制方案的设计都具有基础性的意义。本文以网络论坛中的文章数为研究对象,分析研究文章数序列的特性,提出文章数序列的描述模型,并在此基础上提出网络论坛文章数序列的预测方法。首先,对实际数据进行分析,验证和证明文章数序列的自相似性特性。然后,用短相关特性和长相关特性的时间序列分析方法对自相似网络论坛文章数进行建模分析,给出了利用马尔可夫模型、FARIMA模型进行建模的方法和步骤。为了避免FARIMA模型在结构辨识和参数估计上的复杂性,通过分析ARMA模型的参数估计方法,提出了基于传统ARMA模型的改进模型λ-ARMA。通过对实际数据的实验,验证了FARIMA模型和λ-ARMA模型在文章数序列建模方面的适用性。建立模型的一个重要用途是进行预测,本文结合实际数据给出了预测的算法和步骤,并通过实验验证了不同模型在不同置信区间内的适用性。本文的研究结果表明从文章数的角度分析网络论坛是必要的而且可行的。本文的工作将为下一代网络,尤其是网络论坛、博客等的构建和管理提供参考。更多还原

【Abstract】 With the introduction of the concept Web2.0, high-speed Internet gets more and more attention. Online forums, blogs and other forms of communication have become hot Internet applications. On the one hand, the topic is a major focus of web forum research; the other hand, due to various business decisions (such as product market research or advertising), the forum or blog user participation is becoming a new hotspot, the number of articles on the forum has also recently received attention. Accurate measurement of dynamic changes is essential for analysis, understanding and simulation in online forums, and also has a fundamental significance to guide the design of control scheme.In this paper, the number of articles in Online Forum is to be studied. This paper describes several models. And on that basis, this paper raised the method for forecasting the sequence of the number of articles. First, through analysis of actual data, this paper validates and certificates the self-similarity of article number sequence. Then, the short and long correlation time series analysis methods are applied to the modeling and analysis of self-similarity. This paper gives the modeling method and procedures of Markov models, FARIMA model. In order to avoid the complexity of structure identification and parameter estimation on the FARIMA model,λ-ARMA model is proposed based on the traditional ARMA model. Experiments on real data verify the applicability of FARIMA model and X-ARMA model.An important use of modeling is to predict, this paper gives the prediction algorithm and steps. Through experimental verification of different models, this paper validates the different applicability within the confidence interval.The study results show that from the perspective of the number of articles are necessary and feasible. This work will provide references for the next generation of networks, especially in online forums, blog.更多还原