【作者】 何军；

【导师】 周明天；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2002， 博士

【摘要】 Internet／WWW是一个巨大的、开放的、分布自治式的信息网络系统，为人们提供了内容丰富、数量巨大的信息资源。用户以Browser／Server模式在WWW中漫游，寻找自己所需的信息。随着Internet／WWW的发展，大量的新信息不断地加入Internet／WWW，出现了日趋严重的信息过载现象，使得以传统的Browser／Server模式在信息海洋中找到所需信息成为一项越来越困难的任务；同时大量用户的并发上网也造成了网络拥塞和服务器过载。因此有必要建立新的信息服务方式，它能将有用信息适时送到(尤其是网络空闲时)所需要的用户处，并具有节约用户时间、节省网络带宽、减轻服务器负载以及系统性能受网络规模、用户数影响较小等优点。信息过滤正是具有上述优点的一种技术方案，本文对信息过滤中的一些关键技术进行了研究，得出了如下创新性的成果： ● 建立了基于话题的信息空间与用户兴趣模型。为向用户提供更优质的服务，信息过滤技术将信息与用户信息需求进行比较匹配，为此需要对二者的特征进行描述。已有的基于关键词的信息特征抽取和用户兴趣模型只注重了语义的内涵，缺乏对语义外延的描述能力，不利于为用户提供更个性化的服务。本文从话题的角度，对信息、信息的载体和用户的兴趣进行了分析。通过挖掘用户感兴趣话题的外延特征扩展了传统的用户兴趣模型。给出了以话题为单位的信息存储结构及定位、查询技术，提出了基于活动的用户感兴趣话题的发现算法。 ● 提出了可以高效处理大量user-profile文件的多级索引结构及算法，用背包问题模型来解决给用户提供恰当数量信息的问题。在大规模信息过滤系统中，高效地处理数量非常巨大的user-profile文件成为影响系统性能的重要因素。论文提出了一种基于用户特征的多级索引结构及算法，使进入信息只需与相关的user-profile文件匹配，而无须与所有的user-profile文件进行比较，从而可以极大地减少需要处理的user-profile文件的数量，适用于大规模信息过滤系统。针对人类精力有限的生理特征，建立了用户使用信息的背包问题模型，以免用户被过多不需要的信息所困扰。．提出了一种用于 Internet的三级信息发布体系结构。该体系结构的信息发布分为三级：信源发布、Internet发布、本地网发布，旨在为用户提供更广泛、全面的信息。．给出了一种Internet信息递送的体系结构及各部分的功能。已有的信息过滤体系是基于user十rofile来构建的，难以适应大规模信息递送的需求。本文在此基础上定义了表示用户群兴趣的server巾rofile机制，负责该体系结构中递送服务器之间以及递送服务器与管理服务器之间的管理信息的处理和传送，以提高网络吞吐率和系统效率。并设计了递送服务器在管理服务器的支持下，相互之问进行信息订阅、发布的协作机制。．设计了一种本地网信息发布系统。其任务是由服务器将信息按内容分频道向高速下行信道进行循环发送，用产代理倾听信息流并选择合适的信息提供给用户。论文描述和分析了系统各部分的功能；在综合考虑信息的长度、需求率、己被广播的次数、为再次发送而等待的时间和己存在的时间等特性的基础上，提出了一种多盘广播调度算法，它可以更恰当地将有效信息递送给用户并具有较短的平均访问时间。．设计了一个由用户代理、过滤器、本地信息库等组成的用户端系统。论文描述了用户端系统各部分的功能设计和特点。提出了一种以时间维作为参量、依据在线用户的反馈信息、基于用户过程和系统过程的针对待发送信息的调度算法，以减少信息的冗余发送。设计了一个便于用户使用本地信息的用户界面。更多还原

【Abstract】 Internet/www is a huge, open, distributed and autonomous information network system that provides people rich and numerous information resources. Users surf the WWW, and dig the needed informations by Browser/Server model. With the development of the Internet/WWW, numerous fresh resources continuously join into the Internet/WWW that makes the emerging information-overloading phenomenon more often. It is a tough task to find useful informations from the Information Ocean by B/S model. Simultaneously, numerous users concurrently accessing the same website result in network congestion and server overloading. So it is necessary to innovate a novel information service mode which has the effectives as follows, actively deliverying useful information to target users timely; timesaving and bandwidth-saving; alleviating server payload; relating less between system performance and network scale/user number. Information filtering technology is such a tactic soluton with the effectives mentioned above. This dissertation presents innovative achievements on some key technologies of information filtering as follows,Building the information space model and topic-based user interest model. To provide users better service, information resources should be compared with user demands, and filtered to match those demands with information filtering technology in order to provide much better Quality of Services to users. So, both information and user demands need to be characterized systematically. The existent keywords-based information feature extraction methods and user interest models pay only attention to semantic connotation while ignored utilization of semantic extension. Thereby, it is insufficient for providing users with more personal services. From the viewpoint of regarding topic, the information characters, information media and user interests are analyzed in the dissertation. The traditional user interestmodels are expanded by extending the user-interested-topics. The information storage structure, location and quering technology of information filtering are presented, and an user-interested-topics’ discovering algorithm based on browsing behavior is given.Proposing the multilevel indexing structure and algorithm that can effectively handle the huge user-profile files, and adopts the model of knapsack problem to solve the problem for providing suitable volume of information to users. In massive information filtering system, to effectively handle the huge user-profile files associates closely with system performance. This dissertation presents a multilevel indexing structure and algorithm based on user characteristic. With the algorithm, incoming information need only match with those related user-profile files but all profile files. So user-profile files needed to be process can be greatly reduced which is suitable to the massive information filtering system. Aiming at the physiological characteristic of limited humankind energy, the model of knapsack problem for information consuming is built to assure users away from the unusual redundant information in the dissertation.Proposing a three-tier information publishing architecture for the Internet. The three-tier architecture is designed to provide users much better comprehensive information services which consists of three information publishing steps that is information source publishing, Internet publishing and LAN publishing.Presenting an information delivery system and each subsystem functions used in the Internet. The known information filtering architecture was built based on user-profile which unsatisfies the need of massive information delivery system. Based on it, server-profile mechanism to express interests of any certain user groups is defined in the dissertation, which can be regarded as managing and transferring the informations between different delivery servers or between delivery servers and manage servers in the information delivery architecture. With the tailored mechanism, t更多还原