【作者】 赵银春；

【导师】 朱征宇；

【作者基本信息】 重庆大学 ， 计算机软件与理论， 2004， 硕士

【摘要】 随着信息技术的发展,特别是因特网应用的普及,出现了“信息爆炸”的问题,即信息非常丰富而知识相对匮乏。如何管理WWW上的大量信息,以满足用户不断增长的个性化的信息需求,是研究人员面临的新课题,个性化服务技术已经成为当前信息服务领域的研究热点之一。所谓个性化服务就是指对不同的用户采取不同的服务策略,提供不同的服务内容,其关键在于必须知道用户的兴趣,并准确地建立用户兴趣模型。用户兴趣模型能否准确地反映用户的兴趣决定了系统提供个性化服务质量。本文首先对当前主要的web挖掘技术和用户兴趣建模技术进行了分析,提出了以web用户浏览内容分析为主和以浏览行为分析为辅的用户兴趣挖掘过程模型。然后,对文本页面的表示技术进行了初步的研究和探讨,包括:文本的向量空间模型表示、特征项的选择和抽取算法,将文本页面表示为结构化的向量空间模型格式。接着,本文重点探讨了文本页面的聚类分析和用户兴趣模型的建立两个方面。在分析了现有聚类算法和实际应用环境后,提出了将层次凝聚法(agglomerative algorithm)和平面划分法(K-means algorithm)相结合的新算法:首先利用层次凝聚法进行初始聚类确定初始聚类中心和k值,然后用K-means算法进行聚类分析。在聚类结果的基础上,采用二层树状用户兴趣模型来表示每一个用户的兴趣:表示为(,),(,),…,(,))的加权矢量格式。为了便于用户兴趣模型的使用和更新,每一兴趣类也采用向量空间模型来表示,内容页面与兴趣类的比较就可以采用常用的相似性函数来进行相似度计算。每一兴趣类权值的确定分析了以下三个因素:(1)包含较多的文档页面数量;(2)包含尽可能多的内容页面,尽可能少的辅助页面;(3)尽可能少的页组内相互访问连接。最后,对上述改进方法做了较为全面的模拟实验。根据模拟实验分析表明,本文所改进的聚类算法实现简单,准确率较高;提出的用户兴趣模型能较准确地描述用户兴趣所在,在个性化推荐服务中具有实际应用价值。本文进一步的工作将致力于用户兴趣模型准确性的提高,并将得到的用户兴趣模型应用于个性化系统的推荐服务。更多还原

【Abstract】 With the development of Internet and Information Technology, the problem of “Information Explosion” has arisen , that is, “Rich Data and Poor Information”. How to manage the tremendous mount of information on WWW to meet the growing needs of personalize information is a new subject for our research. Personalization has been the focus of research. Personalization, that is give different service-strategy and different service-content to different user. Knowledge of user interests and describe them by user profiles are the importance. Therefore, the effectiveness of personalize information service provided by the system is determined by the fact whether or not the user profiles reflect user interests exactly.After the studying of the key technologies――web mining technology and modeling user profile, the author suggests the model of mining user interests. The model is based on user viewed content and combining with analysis of user’s behavior. Through analyzing document expressive model、feature extraction and feature weigh value, the web page is been expressed by Vector Space Model.In the paper, the author do hard in two aspect: clustering based on content , creating the user interest model. After the probing into the cluster algorithm existing and the applied practice, the author proposed a new cluster algorithm: combining agglomerative algorithm with K-means algorithm. In the process of cluster, user the agglomerative algorithm get the cluster-means and k firstly, and then use K-means algorithm to do the second cluster. After get the cluster, the author use tree-model to express the user interest, it is as (,),(,),…,(,)). For the sake of using and updating of user interest model, every interest style of user is express by VSM as the web page. So, the compare of web page with user interest style can be valued by similarity function. When calculating the interest style weight, the author consider three factors: (1)the large mount of web page; (2)In web page muster, the content page is more than assistant page; (3)the value of GILD(group inter-link degree) must be little.Finally, the author experiment on the advanced method discussed above. According to the experimentation and analyses, prove that the new cluster algorithm <WP=6>and tree format interest model are reliable, and can be applied in personalization system.Lastly, the deep work of this paper is that developing the validity of user interest model, and applying it into the recommendation.更多还原