An Information Retrieval using Conceptual Index Term for Technical Papers

本文提出了一种基于科技论文语义信息的信息检索方法。该方法适用于数字图书馆（DL）。数字图书馆用户需要检索符合其语义要求的信息。一个重要的问题是减少由单个用户之间的请求差异引起的检索错误。为了解决这个问题，我们使用了一些自然语言处理技术，以及描述单词和概念之间关系的字典。为了提取科技论文的语义信息，我们应用形态学分析程序对科技论文图像中的文本数据作为OCR结果进行分析。在形态分析之后，我们只提取名词，并检查其出现频率的分布，以供以后使用。利用EDR概念词典提取名词概念，并计算其概念出现频率分布。结合词的出现频率分布和概念的出现频率分布，得到与科技论文主题相对应的概念。本研究的核心思想之一是将概念作为包含概念间关系的概念路径来处理。这有助于对技术论文主题的概念进行抽象。实验结果表明了该方法的有效性。我们还实现了一个概念信息检索原型系统。在这篇论文中，我们关注的是用日语写的技术论文。由于EDR词典也有英语单词的概念词典，我们可以很容易地将我们的方法应用到英语技术论文中。EDR概念词典的特点之一是概念的表达与日语词汇和英语词汇具有共性。这将有助于跨语言信息检索。

In this research, we proposed a new Information Retrieval (IR) method using semantic information from technical papers. The proposed method is suitable for use in Digital Libraries (DL). Users of Digital Libraries need to retrieve information which meets their semantic requirements. An important problem is the reduction of retrieval errors caused by differences in requests among individual users. To solve this problem, we used some techniques of natural language processing, and dictionaries that describe the relations between words and concepts.To extract semantic information of technical papers, we applied morphological analysis program for text data derived from images of technical papers as OCR results. After morphological analysis, we extract only nouns and examine their distribution of word appearance frequency for later use. We also extract concepts of nouns with EDR concept dictionary and calculate their distribution of concept appearance frequency. Combining the distribution of word appearance frequency and that of concept appearance frequency, we get the concepts which correspond with the subject of technical papers. One of the key ideas of this research is handling concept as concept paths which include the relations among concepts. This helps abstraction of concepts of the subjects of technical paper. Experimental results show effectiveness of the proposed method. We also realized a prototype system of conceptual information retrieval. We focused on technical papers written in Japanese in this paper. As EDR dictionary also has concept dictionary of English words, we can apply our method to technical papers written in English easily. One of the characteristics of EDR concept dictionary is that expression of concepts are in common with Japanese words and English words. This will help cross lingual information retrieval.