Development of a Full Text Retrieval System based on Logical Structure of Documents

调查人员研究和开发了一个系统，用户可以利用该系统从包含学术论文等全文文件的全文数据库中检索文件。1.在详细分析全文数据库检索请求的基础上，根据SGML的文档类型定义，详细设计了一种用于高效描述检索和显示请求的形式化表示法DQL。2.他们设计了一个用户系统软件，该软件显示文档结构，允许用户使用定点设备选择元素，并指定检索条件和显示说明。该系统在使用图形用户界面系统的工作站上开发。3.他们在一台大型计算机上开发了一个服务器系统软件，该软件将文件存储在关系数据库管理系统中，将文件结构表示为元素之间的二项式关系，并处理用DQL描述的检索请求。4.提出了一种服务器与用户系统之间的通信方法，并通过局域网和综合业务数字网的实验验证了该方法的有效性。5.对服务器系统的整体性能、可用性、检索精度等进行了评估，并针对实际使用中存在的问题和解决方案进行了思考，通过评估，揭示了服务器系统性能提升的必要性。进一步的调查表明，在二项关系记录中纳入参考记录标识是有效的。实现这一方法是未来的问题。还揭示了用户系统更容易理解地表示特定检索请求的必要性。对已提出的几种方法进行了实验研究，验证了该系统的设计思想和实现方法的可行性，提出了解决存在问题的方法，展示了该系统的现实性。

The investigators have conducted research and development of a system with which users can retrieve documents from a full text database containing full documents such as academic papers. Results obtained were as follows: 1. Based on detailed analysis of retrieval requests for full text databases, they made detailed design of a formal notation DQL (Document Query Language) for describing retrieval and display requests efficiently according to the document type definition of SGML. 2. They designed a user system software which displays document structures and let users select elements with pointing devices and specify retrieval conditions and display instructions. The system w as developed on work-stations using graphical user interface systems. 3. They developed a server system software on a main frame computer which stores documents in a relational database management system by expressing the structure as binomial relations among elements and processes retrieval requests described in DQL. 4. They developed a communication method between sever and user systems and confirmed the functionality by experiments using LAN and ISDN. 5. They evaluated the overall performance, usability, retrieval precision and so on and considered on the problems and their solutions toward the practical use.Through evaluation, the necessity of performance improvement of the server system was revealed. Further investigation has made it clear that including referenced record identifiers within the binomial relation records is effective. It is the future issue to implement this method. The necessity for the user system to represent specified retrieval requests more understandably was also revealed. Several methods already proposed should be studied by experiments.As the result of the research, feasibility of the design concept and the implementation method of this system was confirmed, the approach to the existing problems was presented, and the reality of the full-fledged system was shown.