基于多文档摘要的研讨文本分析方法及应用

发布时间：2018-05-17 07:34

本文选题：研讨系统 + 研讨文本　；参考：《湖北工业大学》2017年硕士论文

【摘要】：群体研讨支持系统是综合集成研讨厅的一个重要组成部分,其作用是支持群体成员在线交流自己的意见并对这些意见进行综合分析。在研讨过程中,会产生大量的发言文本信息,如果不对这些文本进行分析处理,则不利于研讨参与者识别不同发言内容之间的逻辑关系。目前,对研讨信息的分析处理多采用人工智能领域中的辩论推理方法,即通过分析发言结点之间的攻击或支持关系及攻击或支持的强度确定发言结点的可接受性或发言结点的共识值。但是,在基于文本的研讨过程中,研讨文本之间没有明确的攻击与支持关系,这时群体共识隐藏于研讨文本内容之中,这就需要对研讨文本内容进行分析处理。针对这个问题,本文提出对多研讨文本内容进行分析处理并提取共识的方法,并将该方法运用于群体研讨支持系统之中。具体工作如下:1、提出一种基于多文档摘要的研讨文本分析方法。该方法首先对研讨文本进行预处理;然后利用启发式聚类算法对预处理后的文本进行聚类分析,形成多个文本簇,并采用多文档摘要算法提取各文本簇摘要;最后采用D3技术展示聚类结果和摘要内容。2、在分析经典的Text Rank算法的基础上,提出一种改进的Text Rank文本摘要算法。该算法的核心思想是将句子位置、关键词句子和句子长度等信息加入到句子权重计算中,同时给出了改进后句子相似度的计算方法并通过实验证明该算法是可行的。3、开发一个基于Web的研讨文本分析子系统,并把该子系统整合到实际的研讨系统中。用一个研讨实例中的研讨文本集进行实验分析,通过调节聚类算法中的阈值,可以得到不同的聚类结果和文本簇的摘要。实验证明,本文所提出的基于多文档摘要研讨文本分析方法是可行的,能有效促进专家思维收敛,提高研讨效率。
[Abstract]:Group discussion support system is an important part of the integrated research hall. Its function is to support group members to exchange their opinions online and analyze them synthetically. In the process of discussion, a large amount of speech text information will be produced. If these texts are not analyzed and processed, it will be unfavorable for the participants to identify the logical relationship between different speeches. At present, the reasoning method in the field of artificial intelligence is mostly used in the analysis and processing of the discussion information. In other words, the acceptability of the speaker node or the consensus value of the speaker node are determined by analyzing the attack or support relationship between the speakers and the intensity of the attack or support. However, in the process of text-based research, there is no clear attack and support relationship between the study texts, so the consensus is hidden in the content of the research text, which needs to be analyzed and processed. In order to solve this problem, this paper puts forward a method of analyzing and processing the contents of multi-discussion texts and extracting consensus, and applies this method to the group discussion support system. The main work is as follows: 1. A research text analysis method based on multi-document summary is proposed. The method firstly preprocesses the research text, then uses the heuristic clustering algorithm to cluster the preprocessed text to form multiple text clusters, and uses the multi-document summarization algorithm to extract each text cluster summary. Finally, D3 technique is used to show the clustering results and summary content. Based on the analysis of the classical Text Rank algorithm, an improved Text Rank text summary algorithm is proposed. The key idea of the algorithm is to add sentence position, key word sentence and sentence length into sentence weight calculation. At the same time, the improved method of sentence similarity calculation is given, and the algorithm is proved to be feasible by experiments. A discussion text analysis subsystem based on Web is developed, and the subsystem is integrated into the actual research system. The experimental analysis is carried out with a discussion text set in an example. By adjusting the threshold in the clustering algorithm, different clustering results and abstracts of text clusters can be obtained. The experiments show that the method proposed in this paper is feasible and can effectively promote the convergence of expert thinking and improve the efficiency of research.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【参考文献】