Study on Information Utilization System for Heterogeneous Contents

该项目旨在开发利用异质内容的技术。我们研究了Web的链接和结构分析、跨媒体处理技术、Web的认识论框架，并开发了评估Web信息利用方法的语料库1)提出了一种利用Web的文本和图形结构的信息提取和组织方法-基于链接结构的Web页面聚类方法-基于非线性时间-内容分析的主题跟踪2)我们提出了一些先进的多媒体处理和利用方法如下：聚焦于媒体异构性：-从多语言文本集合中进行主题检测-基于内容类型的用户自适应文本摘要-通过增强基于内容的基于注释的图像检索模型来进行跨媒体搜索-Junii+：图像检索的用户界面-利用访谈视频档案进行学习3)我们组织了一系列评估研讨会(NTCIR)，多名研究人员参与开发了新的测试床，每个测试台都包含一个用于研究异质数字内容的通用测试数据。作为结果，例如，我们通过爬行-JP域建立了TB级的数据集，并建立了符合实际情况的评估方法。我们分析了工程师处理和建模Web信息源的认识论框架，并将其与现代印刷书籍系统进行了对比。在分析的基础上，我们得出的结论是，很难直接应用图书馆理想中所代表的现代信息积累的精髓概念所定义的模式，并指出要充分挖掘网络信息源的潜力，“信息编辑”是必要的。

This project aims at developing technology for utilizing the heterogeneous contents. We studied link and structural analysis of Webs, cross-media processing technology, epistemological framework of the Web and developed corpora for evaluating information utilization methods for the Web.1) We developed an information extraction and organization methods using the textual and graphical structure of the Web-Web page clustering methods based on the link structure-Topic tracking using non-linear time-content analysis2) We proposed some advanced methods for processing and utilizing multimedia as follows, focusing on media heterogeneity:-topic detection from multilingual text collection-user adaptive text summarization based on content types-crossmedia search by enhancing annotation-based image retrieval model with content-based features-JuNii+: user interface for image retrieval-utilizing interview video archives for learning3) We organized a series of evaluation workshops "NTCIR", in which a number of researchers participated to develop new testbeds, each of which consists of a common test data for research on heterogeneous digital content. As the results, for instance, we built up a terabyte-scale dataset by crawling the -jp domain, and established evaluation methodologies to meet the practical situation. These contributed to the progress of the research in this area4) We analyzed the epistemological framework within which engineers process and model the Web information sources, contrasting it with the modern system of printed books. On the basis of the analysis, we concluded that it is hard to directly apply the model defined by the quintessentially modern concept of information accumulation as represented in the ideal of libraries, and showed that "information editing" would be necessary to explore fully the potential of web information sources.