A Partial match technique for multi-attribute keys and its text retrieval application

关键词的高效提取是文本检索系统中的重要任务。在日语文本中，有许多由某些类型的字符组成的复合词（片假名、汉字等），并且文本中词与词之间没有分隔符。因此，从这样的文本中提取关键字需要花费大量时间。本文提出了一种从复合关键字中检测关键字的技术，该技术通过引入一组规则来表示关键字构造的多属性条件。使用有限数量模式的字符串模式匹配机对规则进行匹配，并将关键词候选词与长词和短词信息一起存储。通过理论分析对该方法进行了估计。通过对34个日文文本文件的仿真结果表明，该算法的提取效率达到19.4ms/KB，期望关键词的提取率比传统方法有所提高。

Extracting keywords efficiently is an important task in text retrieval systems. In Japanese text, there are many compound words consisting some kinds of characters (Katakana, Kanji, etc.) and the text has no delimiter among words. Therefore, extracting keywords from such a text takes a lot of time. This research presents a technique of detecting keywords from compound keywords by introducing a set of rules, which represents multi-attribute conditions for keywords conctruction. A string pattern matching machine for a finit number of patterns is applied to matching of the rules and storing keyword candidates together with information bout both long term and short term words. The approach is estimated by theoretical analysis. By the simulation results for 34 Japanese text files, it has been that the algorithm presented has performed 19.4ms/KB and that the ratio of extracting expected keywords increase from the traditional approaches.