A study on multi-Iingual information retrieval with structured index

在本研究中，我们探讨了一种新的信息检索方法的潜力，它被命名为“结构化索引”。“在这种方法中，我们生成一个以二叉树结构表示的索引，该索引是通过对组成科学论文标题的单词之间的依赖关系进行分析而创建的。这种方法有望优于传统的基于关键字的信息检索方法，因为这种索引将更适合于匹配嵌入在用户的查询的意图。此外，由于索引更面向概念，该方法更适合于跨语言信息检索。首先，我们制作了一个基本的日文论文标题词法分析和词与词之间依存关系分析的软件系统。基于这些分析，我们研究了（1）适合于索引结构的方法，以及（2）检索处理的通用算法。经过这些前期工作，我们制作了一个实用的检索软件 ...更多信息 该系统被应用于日本最大的测试集合之一，NTCIR。我们为标题和摘要字段建立了一个结构化的索引，评价表明，我们的新方法优于传统的方法。我们还设计了一个方法，将我们的方法应用到英语信息检索。在跨语言信息检索方面，我们同时提出的另一种方法“相关叠加（RS）模型”表现出更好的性能，我们决定选择RS模型进行跨语言信息检索。在测试集上的实验表明，我们的系统在日语和英语集合中取得了更好的检索性能。我们所实现的软件由两部分组成，一是软件的开发。例如，语言独立部分和语言依赖部分。这种结构更适合于进一步开发其他语言的功能增强。我们还制作了一个基于Web的信息检索用户界面，用于展示我们的研究成果。少

In this research, we have investigated the potentiality of a novel information retrieval method which is named "Structured Index." In this method, we generate an index represented in a binary-tree structure, which is created through a dependency analysis between words that compose titles of scientific papers. This method is expected to outperform conventional keyword-based information retrieval methods, because this index would be more appropriate for matching to the intention embedded in users' queries. Furthermore, this method can be more suitable for cross-lingual information retrieval since index is more concept-oriented. Firstly, we made a fundamental software system with Japanese language morphological analysis of paper titles and dependency analysis between words. Based on these analyzes, we investigated (1) a method appropriate for index structuring, and (2) a general algorithm for retrieval processing. After these preliminary works, we have made a practical retrieval software … More system which was applied to one of the largest Japanese test collections, NTCIR. We made a structured index for title and abstract fields and the evaluation has shown that our new method outperforms conventional methods. We also designed an approach to apply our method to English information retrieval. In terms of cross-lingual information retrieval, another method "Relevance Superimposition (RS) Model" that we also have proposed concurrently has shown better performance, and we decided to choose RS model for our cross-lingual information retrieval. Experiments on the test collections have shown that our system achieved better retrieval performance in both Japanese and English collections. The software we have implemented is composed of two parts, i. e., the language-independent part and the language-dependent part. This structure is better for further development of functionality enhancement for other languages. We also made a web-base user interface of information retrieval for the demonstration of our research achievements. Less