个性化信息服务中的用户兴趣迁移研究

发布时间：2018-02-13 08:40

本文关键词： 兴趣迁移兴趣收集用户聚类遗忘函数时间窗　出处：《电子科技大学》2009年硕士论文　论文类型：学位论文

【摘要】：当今社会是信息爆炸的社会,对此,人们早已没有争议。为了能有效地利用呈指数级增长的海量数据,而不至于淹没在信息的海洋中,人们开始对数据挖掘技术进行深入研究,并使之成为计算机领域的一个研究热点。数据挖掘研究的是如何在多样的、海量的数据中提取出蕴含的、有用的、潜在的,但不容易被发现的知识和模式。由于网络的普及与发展,出现了新问题,就是由于信息太多,要么难以及时地消化吸收,要么难以有效地使用,即所谓“信息过载”和“信息迷航”的问题。人的接受能力和认知能力毕竟有限,时间精力更是有限。所以,一种普遍的现象就是网络用户面对铺天盖地的资讯只能被动接受,而在寻找自己感兴趣的信息时又如同大海捞针。自己不感兴趣的常常躲不掉,自己感兴趣的往往找起来又很困难。于是新的需求产生了,用户希望能及时获得自己感兴趣的信息,能有满足自己个性化要求的服务。新的需求就是新的市场,信息服务提供商也迫切想开发出针对用户个性化需求,满足用户兴趣爱好的信息服务系统。从而增加竞争优势,拓展盈利空间。两方面的愿望,促成了个性化信息服务的诞生。个性化信息服务的关键和基础就是收集用户兴趣,建立用户兴趣模型,掌握用户兴趣的迁移变化。知道了用户的兴趣,才知道该如何“投其所好”,有的放矢的开展个性化服务,比如有针对性的改进站点结构和设计,推荐信息,定制广告等。本文完整介绍了基于Web的用户个性化兴趣挖掘过程,并对用户兴趣收集、用户兴趣建模、用户聚类和用户兴趣迁移做了重点研究。针对用户兴趣收集提出了基于用户隐式兴趣反馈和显式兴趣反馈相结合的用户兴趣收集模型。在用户兴趣建模阶段,将用户兴趣分为长期兴趣和短期兴趣进行详细分析,从而提出相应的算法。随着时间的推移,一些原有的兴趣会过时,一些新的兴趣会产生,需要对用户已有兴趣进行更新。本文又重点研究了用户兴趣迁移,就是研究用户兴趣随时间变化后如何及时更新兴趣,如何淘汰旧兴趣添入新兴趣。以往的研究通常有时间窗口法,渐近遗忘法等。本文将多种方法相结合,提出了采用混合模型进行建模的算法,这能准确的反映用户兴趣迁移变化。
[Abstract]:Today's society is a society of information explosion, and there has been no dispute about it. In order to effectively utilize the massive data with exponential growth, and not to be submerged in the ocean of information, people began to study the technology of data mining in depth. Data mining is how to extract the knowledge and patterns contained, useful, potential, but not easy to find in the diverse and massive data. As a result of the popularity and development of the network, new problems have arisen, that is, because of too much information, it is either difficult to digest and absorb in time or to use it effectively. That is, the so-called "information overload" and "information confusion". After all, people's ability to accept and recognize is limited, and their time and energy are even more limited. Therefore, it is a common phenomenon that network users can only passively accept the overwhelming amount of information. And when you look for information you're interested in, it's like looking for a needle in a haystack. You can't hide what you're not interested in. It's often difficult to find what you're interested in. Users hope to get the information they are interested in in time and to have services that meet their own personalized requirements. The new demand is a new market, and the information service providers are eager to develop personalized requirements for users. The information service system to satisfy the interests and interests of the user, thus increasing the competitive advantage and expanding the profit space. The desire of two aspects has contributed to the birth of the individualized information service. The key and foundation of the personalized information service is to collect the interest of the user. Set up user interest model, master the migration change of user interest. Know the user's interest, know how to "give it what you like", carry out individualized service in a targeted way, such as improving the structure and design of the site, recommending information, Custom advertising, etc. In this paper, the process of user personalized interest mining based on Web is introduced, and the collection of user interest, the modeling of user interest, This paper focuses on user clustering and user interest transfer. A user interest collection model based on implicit user interest feedback and explicit interest feedback is proposed for user interest collection. In the stage of user interest modeling, a user interest collection model based on implicit interest feedback and explicit interest feedback is proposed. The user's interest is divided into long-term interest and short-term interest for detailed analysis, so that the corresponding algorithm. Over time, some of the old interest will become obsolete, some new interest will be generated. This paper also focuses on user interest migration, that is, how to update user interest timely after the change of interest over time. How to eliminate old interests and add new ones. The previous researches usually include time window method, asymptotic forgetting method and so on. In this paper, a hybrid model modeling algorithm is proposed by combining many methods. This can accurately reflect user interest migration changes.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：TP311.13

【引证文献】