The Research Of Technology And Evaluation On Meeting Summary Extraction

The development of information technology is so fast, people can achieve information from different channels, how to get information that needed from huge amounts of information for users has become a hot topic in today’s information field of research. Today, a lot of meetings take happens, and as a result of the meeting itself characteristic is easy to appear some topic which has nothing to do with the conference contents, if people take the time to attend the meeting or browse the entire content will waste much time, people can get through browsing the meeting at this time of the effective information, not only save a lengthy meeting time, but also improve the efficiency of the information access which caused the wide attention of the researchers recently.This article first introduced the voice meeting the extraction technology including supervised learning methods and unsupervised learning methods.Supervised learning method has been widely used in the speech summary extraction. Using a supervised learning method, the extraction was seen as a binary classification task to decide whether a sentence is the summary sentence.Support vector machine classification method is widely used at present, the SVM in many binary classification task has good performance. But those closeto the classification of the sample itself does not have division of the sentence with the obvious characteristics of the sentence, although there are size to distinguish the confidence values of the samples, but it is still in the classification of a fuzzy zone, there is no obvious priority between each other.Then, using the Maximal Marginal Relevance method of SVM for the post-processing, and through the experiment results show that the MMR method can not only remove the redundant information in the paper, and compared with pure using SVM classification performance is higher in the case of the extraction.Usually the meeting is based on the characteristics of a particular topic, for the purpose of obtaining high quality meeting summary, based on the text recorded as treatment object and the SVM and the MMR extraction algorithm,aiming at the current meeting around a topic to discuss, thus put forward a new meeting summarization technology confusion with SVM and MMR based on the theme. This method is based on the topic keywords scores and takes into account the sentence rating information such as position characteristics and importance of ROUGE value evaluation method is used to evaluate the performance of the summary extraction. Experimental results show that the SVM method combined with the MMR method based on the theme is better.Previous studies which focused on meeting summary extraction algorithm have had a lot of improvement methods and different evaluation methods. Most of them is meeting the improved extraction algorithm and use, such as the use ofa separate supervision, the extraction method, or use the unsupervised extraction method, or half supervision and the extraction method, and some researchers combine many kinds of algorithm for the extraction. The extraction method based on the theme of the meeting, mainly along a meeting in one or several specific topics for the extraction of the sentence, so the extracted in the big meeting in the text showed high performance and advantages, easier for the user to read and understand. Through MATLAB experiments show that the improved method has good performance in meeting the extraction system.