Producing and Evaluating Encyclopedic Content by Reorganizing Heterogeneous Information

我们提出了一种从万维网中自动抽取术语描述的方法，并建立了一个称为“旋风”(http://cycbne.slis.tsukuba.ac.jp)，)的网络搜索站点，在这里用户可以高效地获取针对特定词义的百科全书式的术语描述。大约有75万个日语词汇被编入了词条索引。然而，为了解释某些词头词，特别是与诸如设备和动物之类的实体相关的那些词头词，除了文本描述之外，呈现实体的图片是有用的。然而，由于人工编写的局限性，现有的百科全书往往缺乏新的术语和现有术语的新定义，针对上述问题，本研究的目的是产生百科全书内容，为此，我们对万维网和电视广播中的异质信息进行了重组。我们提出了一种将Web上的图像和Cyclone中的文本描述进行集成的方法。我们的方法通过文本分析来解决图像含义上的任何歧义，从而根据词义对多义词的图像进行分类，例如“HUB(网络设备和车轮中心)”。结合信息检索和语音识别技术，提出了一种文本和视频信息关联的方法。此外，为了关联跨语言的信息，我们提出了词汇化和音译方法。我们的研究是朝着多媒体百科全书的自动编译迈出的一步。

We proposed an automatic method to extract term descriptions from the World Wide Web and have built a Web search site called "Cyclone" (http://cycbne.slis.tsukuba.ac.jp), where users can efficiently obtain encyclopedic term descriptions fir specific word Senses. Approximately 750, 000 Japanese terms have been indexed as headwords. However, to explain certain headwords, specifically those related to entities such as devices and animals, it is useful to present a picture of the entity, in addition to a textual descriptionHand-crafted multimedia encyclopedias, such as Encarta, integrate ext, sound, usage, and video data to describe a single headword from different perspectives. However; due to the limitations of manual compilation, existing encyclopedias often lack new terms and new definitions for existing terms.In view of the above problem, the objective of this research was to produce encyclopedic content, for which we reorganized heterogeneous information in the World Wide Web and TV broadcasting. We proposed a method for integrating images on the Web and textual descriptions in Cyclone. Our method resolves any ambiguity in the meaning of an image by text analysis, so that images for a polysemous word, such as "hub (network device and center of wheel)", are classified using word senses. We also proposed a method to associate text and video information, for which we integrated information retrieval and speech recognition technologies. In addition, to associate information across languages, we proposed lemmatization and transliteration methods. Our research is a step toward the automatic compilation of multimedia encyclopedias.