基于数据集特征的伪相关反馈中平衡参数自调节方法研究

发布时间：2018-03-13 02:28

本文选题：信息检索　切入点：伪相关反馈　出处：《天津大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着互联网技术的迅速发展,网上文本数量呈指数级增长,如何有效检索这些海量信息成为当前重要的研究课题。文本信息检索是处理海量文本的重要手段。文本信息检索是指从大量文档集合中找到与给定的查询请求相关的、恰当数目的文档子集。面对庞大的文本信息,各种检索模型被相继提出,而其中最有效的一种就是伪相关反馈(PRF)模型。伪相关反馈已经被证明是一种有效提高文本检索性能的方法,并且被广泛使用。对于伪相关反馈来讲,如何选择原始查询与扩展查询词之间的平衡参数是一项重要但却困难的工作。传统方法中,平衡参数在不同的数据集和不同的查询词中通常被设置成相同的经验值。然而由于数据集之间、查询词之间都各不相同,为了提升检索效果,该平衡参数应该有所不同。近期有研究提出了基于机器学习的方法,通过分析基于查询词和反馈文档的特征来预测该平衡参数。本文中,考虑到数据集的差异同样会影响到平衡参数的选择,我们提出了基于数据集的三类特征,包括查询词的信息含量,反馈文档和扩展词的可靠性。本文基于上述三大类特征,通过逻辑回归和交叉验证的方法,自适应的预测伪相关反馈中的平衡参数。首先分析了研究背景意义、国内外研究现状以及主要探究内容;接着介绍了文本信息检索、检索模型等信息检索的相关知识;之后详细介绍了伪相关反馈的相关技术;在详细描述了三大类特征之后,介绍了整个实验的流程以及实验结果。实验结果表明我们提出的方法与之前的研究相比,在提高检索性能方面有着一定的优势。
[Abstract]:With the rapid development of Internet technology, the number of online texts has increased exponentially. How to effectively retrieve these massive information has become an important research topic at present. Text information retrieval is an important means to deal with massive text. Text information retrieval refers to finding out from a large number of document sets related to a given query request. Appropriate number of document subsets. In the face of large text information, various retrieval models have been proposed one after another. One of the most effective is the pseudo-correlation feedback (PRF) model, which has been proved to be an effective method to improve the performance of text retrieval and is widely used. How to select the balance parameters between the original query and the extended query is an important but difficult task. The balance parameter is usually set to the same empirical value in different data sets and different query terms. However, because the query terms differ from one data set to another, in order to improve the retrieval effect, This balance parameter should be different. Recently, a machine learning based approach has been proposed to predict the balance parameter by analyzing the features of query words and feedback documents. Considering that differences in data sets also affect the selection of equilibrium parameters, we propose three types of features based on data sets, including the information content of query words, the reliability of feedback documents and extension words. Through the methods of logical regression and cross validation, the equilibrium parameters in pseudo-correlation feedback are predicted adaptively. Firstly, the significance of the research background, the current research situation at home and abroad and the main contents of the research are analyzed, and then the text information retrieval is introduced. The related knowledge of information retrieval, such as retrieval model, and the related technology of pseudo-correlation feedback are introduced in detail. After describing the three kinds of features in detail, The experimental results show that the proposed method has some advantages in improving retrieval performance compared with previous research.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】