基于纯位置索引高效文档检索.doc

本科生毕业论文 题目：(中文) 基于纯位置索引的 (英文) Efficiency Document Retrieval with Flat Position Index 姓 名： 学 号： 院 系： 信息科学技术学院 专 业： 计算机科学与技术 指导教师： 二○一一年十月一日 摘要 查询语句中的短语查询在现代搜索引擎中占据着较大比例。本文引入一种新的索引结构——纯位置索引（flat position index）结构去处理短语查询。首先，本文对纯位置索引结构及其实现进行介绍；同时，本文提出了一种新的结构（CSLT）来辅助纯位置索引提高其检索效率；并且与在处理短语查询上进行比较实验，纯位置索引在倒排索引文件的空间消耗上会有约6.5%的减少，在短语查询上约有18%的查询速度的提升。此外，本文时下流行的一个国外微博客服务的网站推特（Twitter），根据推特对发布的文本信息（tweet）的最大字节数的限制来修改纯位置索引的索引结构，推特上与索引在布尔查询、短语查询、邻近查询三种查询方式下进行检索的时间性能比较，修改后的纯位置索引可以在涉及位置信息的查询（如短语查询和邻近查询）获得平均60%的查询速度的提升。基于实验结果，本文得出纯位置索引在处理涉及位置信息的查询具有高效性的结论。 关键词：纯位置索引 查询性能 推特检索 Abstract A large proportion of modern search engine queries contain phrases, namely a sequence of adjacent words. In this paper, we use a flat position index to make phrase queries evaluation more efficient. This paper presents a brief introduction of flat position index. In addition, a new structure called CSLT (cache sensitive look-up table) is proposed to assist to improve the performance of query processing. Experiments with GOV2 data collection show that flat position index can reduce the index size by almost 6.5% and speed up phrase querying about 18% compared with word-level index. Furthermore, this paper regards Twitter as a typical example of a collection with short documents, modifying flat position index based on Twitter’s main feature, say a limitation on tweet’s maximum length. Experiments with Edinburgh Twitter Corpus are implemented under three different kinds of query modes, including boolean query, phrase query and proximity query. Compared with word-level index, the modified flat position index achieves nearly 60% speed up for last two query modes, as well as 29% index size reduction. According to the two experiments, this paper shows that flat position index can achieve a better performance in processing queries with position demand, e.