基于Markov网络及laplacian映射的快速相似性检索方法  

FAST SIMILARITY RETRIEVAL BASED ON MARKOV NETWORK AND LAPLACIAN EIGENMAP

作　　者：文辉[1] 王明文[1] 吴水秀[1] 万剑怡[1] 

机构地区：[1]江西师范大学计算机信息工程学院,江西南昌330022

出　　处：《计算机应用与软件》2012年第8期37-40,125,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(60963014;61163006)

摘　　要：快速相似性检索技术对于各种信息检索应用都具有很大的意义,其中基于语义哈希的快速相似性检索即是一个合理有效的检索方式,其检索模型能够在保证语义相关的基础上将高维空间中大量相关的文档数据,映射在低维空间中。虽然近年来许多关于语义哈希的研究都表现了不错的实验结果,但是都没有考虑到利用文档集合自身的信息来加强文档间的相关信息。为了有效利用文档自身信息,提出结合强化文档间邻接关系的马尔可夫迁移过程及使用保留局部信息的拉普拉斯映射方法的相似性检索方式。Fast similarity retrieval techniques are of great significance to various information retrieval applications. Among them the fast similarity retrieval based on semantic hashing is a reasonable and effective retrieval approach, its retrieval model is able tomap a large num- ber of documents data in high-dimension space onto low-dimension space on the basis of ensuring the semantic relativity. Though in recent years many studies in regard to semantic hashing make good results in experiments, however all of them still not take into account to enhance corresponding relations between the documents using the information of the document set its own. Therefore in order to effectively make use of the document information its own, we propose a similarity retrieval approach which combines the Markov transmission process intensifying the adjacency relation between the documents with the Laplacian Eigenmap using reserved local information.

关 键 词：文档检索 语义哈希 马尔可夫网络 拉普拉斯特征映射 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]