【作者】 曲建华；

【导师】 刘希玉；

【作者基本信息】 山东师范大学 ， 管理科学与工程， 2003， 硕士

【摘要】 随着Internet的飞速发展，网络上的信息呈指数增长，其内容之丰富，种类之繁多，堪称世界上最大的信息资源，如何能够更有效、更准确地找到自己感兴趣的内容，滤除与自己的需求无关的信息，真正做到“各取所需”，关系到我们能否充分利用这个巨大的信息资源的问题，这已成为基于Internet的网络信息获取的热点问题，也是本文的研究目标。 近年来，在信息查询领域中兴起了信息过滤(information filtering)技术，它是一种系统化的方法，用来从动态的信息流中抽取出符合用户个性化需求的信息。相比于传统的信息检索模式，信息过滤技术具有较高的可扩展性，能适应大规模用户群和海量信息；可以为用户提供及时、个性化的信息服务；具有了一定的智能和较高的自动化程度。 文本将介绍Web上的信息过滤的基本问题，包括：信息过滤的体系结构，过滤系统的特点及分类，检索和过滤的关系，过滤系统常用的模型以及过滤系统的性能评价指标等，从总体上对过滤问题进行一定的讲述，指出现存的信息过滤系统中存在的问题。 针对过滤系统中存在的部分问题，在现有的信息过滤技术的基础上，结合机器学习中的神经网络方法和智能体Agent的技术，对现有的过滤算法进行了一定的改进，并给出了具有智能性、主动性和扩充性的基于多Agents的个性化过滤系统的模型。本文的主要研究内容为以下三个部分：个性化模式库的表示、建立和维护；改进的过滤匹配算法；基于多Agents的智能过滤系统模型。 对于个性化模式库的表示，本文给出了一种多用户多主题的分类表示方式，使得每个Profile文件只表达一个用户的一个主题，可以更清晰的表达用户的兴趣；对于个性化模式库的建立，本文采用了机器学习中的Hopfield神经网络模型，Hopfield网络具有丰富的联想记忆功能，可以用来对用户兴趣进行联想，建立用户的初始个性化模式库；对于个性化模式库的维护，采用了基于用户反馈的学习方法。 在改进的过滤匹配算法当中，本文结合了布尔模型和向量空间模型的优点，综合考虑了过滤过程中的匹配度和相似度，给出了新的过滤匹配算法，并通过对比模拟实 山东师范大学硕士学位论文验证明了算法的有效性。 结合智能Agent的特性，本文提出了基于多Agents的个性化智能过滤系统模型，对系统中Agent的关系、功能、结构以及系统实现的关键技术和流程进行了详细介绍。该模型充分利用了Agent智能体的特性，将从智能性、主动性、扩充性、易维护性等方面弥补了现有信息过滤系统中的不足，提高速度和精度，帮助人们最大限度的发现自己感兴趣的问题。更多还原

【Abstract】 With the rapid development of Internet, the information on it increases in exponent. It has so huge contents and so many sorts that it may be the largest information resources in the world. How to search out ones’ most interested contents and filter out those information demands irrelevant to users decides on whether we can utilize the huge information resources. It has been the focus questions of searching information on the Web, and it is also the research goal of our article.In recent years, information filtering technology, for short, IF, appears in theinformation searching field. It is a kind of systematic method, and can be used to extract individuated information from dynamic data flows to meet user demands. Contrasted to traditional information retrieval pattern, IF technology has higher extensibility and it can fit for large scale of users and huge information, and can provide timely and individuated information services for users. IF has intelligence to a certain extent and higher roboticized degree.This paper introduces the main issues of information filtering including the system structure of information filtering, the characteristic and classification of filtration system, the relation between retrieving and filtering, the common model of filtering system, the evaluation index of the system performance and so on. It dwells on the IF questions as a whole and points out the existent problems in current IF systems.Aiming at the existing problems in current IF systems and based on the existent IF technology, this article combines the neural network method and intelligent agent technology in machine learning field to improve on the filtering algorithm and give the individuated filtering system model which have the intelligent characteristic, subjective ability and extensibility based on multi-agents. The main research issues of this article are as follows: the representation, creation and maintenance of the individuated pattern-base, the improvedthe improved filtering-matching algorithm and the intelligent filtering system model based on multi-agents.As for the representation of the individuated pattern-base, we introduces a new classification representation method based on multi-users and multi-topics, so as to make each profile only denote one user’s one topic. This method makes it possible to express explicitly the user’s interest. As for the creation of the individuated pattern-base, we adopt the Hopfield neural network model, which has the function of ample association and remembrance and may be used to associate with the user’s interest to create the initial individuated pattern-base. And as for the maintenance of it, we use the study method based on user’s feedback.In the improved filtering-matching algorithm, we put forward a newfiltering-matching algorithm. This method combines the advantages of Boolean model andvector space model and thinks over the matching degree and similarity degree in the wholeof filtering process. The validity of the algorithm has been proved by simulationexperiments.Combining to the characteristics of the intelligent agent, we advance the individuated filtering system model based on multi-agents and give detailed description of the model including the relation between agents, the function, the structure and the key technology to the implementation of the system as well as the implementation process. Making full use of the characteristics of intelligent agent, this model makes up for the shortcomings of the current information filtering system from intelligence, initiative, extensibility and easy-maintenance. It improves on the efficiency and precision of the whole system and can try its best to help us search out the interesting contents on the Web.更多还原