关键词: 自动文摘, 句群, 主题模型, 聚类

Abstract: At present, sentences or paragraphs are considered as a processing unit in most automatic abstracting models. In this paper, an automatic abstracting method was proposed based on sentences grouping. This method adapts an automatic Chinese sentences grouping theory based on MDA. The obtained sentences groups contained better semantic information which was more suitable as a processing unit in automatic abstracting. At the same time, one text was represented as a vector matrix by using the LDA topic model and clustering operation was processed using K-Means algorithm. Then the candidate abstract was generated from clustering results according to some proportions. Finally the obtained abstracts were evaluated by Kappa statistics and Kendall related coefficient. The experimental results show that the overall Kappa value reaches 0.7 and the Kendall related coefficient is more than 0.8, which are all higher than those of the respective good grades by using the method. So the automatic abstracting based on sentences grouping can generate better results compared with the traditional methods which considers sentences as processing granularity.

Key words: Automatic Abstracting, Sentences Grouping, Topic Model, Clustering