Research on the Organization of Heterogeneous WWW Information Space based on Meaning Understanding of Web Pages

万维网（World Wide Web， WWW）是一个异构的、由超链接连接而成的巨大信息空间，是一种重要的信息基础设施。搜索引擎是使用WWW不可或缺的工具；然而，对其他高级支持功能存在强烈需求。在本研究中，我们基于对网页的理解，开发了三个系统来提高WWW空间的实用性和全面性。首先是Fish View系统，它支持每个用户基于用户依赖的视点对从WWW获取的大量文本数据进行整理，并允许他/她高效地读取这些文本数据。该系统的主要特点是一种基于电子词典概念层次的动态词向量文本匹配机制——鱼眼匹配。第二个系统是Web Beholder，它检测指定Web页面及其相关页面上的有意义的更改，通过电子邮件通知用户更改的发生，并在需要时以可视方式显示更改的部分。虽然万维网主要提供股票类信息的全球共享功能，但总的来说，新出现的信息比其他股票类信息更有价值。这个Web Beholder系统允许提供一个新的功能，即在WWW空间中立即传输现在类型的新信息。第三种是Are View系统，它从某个（学术）领域（如人工智能、分子生物学等）的众多（数百或数千）网页中提取少量（10 ~ 30）有用的网页，称为核心网页。核心页面主要是通过从其他页面中引用链接的数量来提取的。系统还可以通过将一个区域中的许多Web页面关联到一个核心页面来对混乱的Web空间进行弱组织。

The World Wide Web (WWW), which is heterogeneous but constitutes a huge information space connected via hyperlinks, becomes an important information infrastructure. Search engines are indispensable tools for the use of the WWW ; yet there exist strong needs for other advanced support functions. In this research, we have developed three systems for enhancing the utility and comprehensiveness of the WWW space based on the understanding of Web pages.The first one is Fish View system, which support each user to sort out many text data obtained from the WWW based on a user-dependent viewpoint and allows him/her to read those text data efficiently. The main feature of this system is a text matching mechanism called Fish Eye matching using dynamic word vectors based on the concept hierarchy of an electronic dictionary. The second system is Web Beholder, which detects meaningful changes on designated Web pages and their related pages, informs the occurrence of the change to users via email, and displays the changed parts visually if requested. While the WWW mainly provides the function of sharing stock-type information globally, newly appeared information is more valuable than other stock parts in general. This Web Beholder system allows to provide a new function of immediately transmitting now-type new information in the WWW space. The third one is Are View system, which extracts a small number (10〜30) of useful Web pages called core pages among many (hundreds or thousands) Web pages in a certain (academic) area such as artificial intelligence, molecular biology, etc. The core pages are extracted primarily using the number of referred links from others. The system can also weakly organize the messy Web space by associating many Web pages in an area to one of the core pages.