摘要： 分布式文本检索系统难以兼顾高效率的数据检索和低成本的索引维护。为此，提出一种基于计数型布隆过滤器的文本检索模型CBFTRM。该模型将物理节点分为数据节点和索引节点，分别采用结构化P2P进行网络覆盖。每个数据节点负责存储文档数据并维护与之相应的倒排索引，同时通过倒排索引中的关键词集合计算出计数型布隆过滤器值，发送给相应的索引节点。每个索引节点建立一棵以部分数据节点的特征信息(包括过滤器值)为叶节点、以过滤器值运算结果为内部节点的搜索树，并在叶节点发生变化时对搜索树进行维护。仿真实验结果表明，该模型文档定位快，索引维护通信量小，而且具有较高的查准率。

关键词: 计数型布隆过滤器, 搜索树, 结构化P2P, 文本检索, 倒排索引

Abstract: The distributed text retrieval system is difficult to take both high retrieval efficiency and low cost of index maintenance into account, so this paper proposes a Text Retrieval Model based on Counting Bloom Filter(CBFTRM) to solve the problems above. This model divides the physical node into the data node and the index node, both of which are overlaid with structured P2P network. Each data node is responsible for storing documents, and maintaining the inverted index of the documents. It also transmits the values of Counting Bloom Filter(CBF) which are computed by the inverted index’s keywords to the corresponding index node. Each index node builds a search tree and maintains it when the tree’s leaf node changes. The search tree is built by leaf nodes with the data node’s character(including their counting bloom filter’s value), and its internal nodes with the result computed by the values of counting bloom filter. Simulation result shows that this model locates the document faster, and has less traffic doing index maintenance and higher precision.

Key words: Counting Bloom Filter(CBF), search tree, structured P2P, text retrieval, inverted index

中图分类号: 

• TP311.13