关于自动文摘系统中文摘句式的一种机器学习方法

A Machine Learning Algorithm of Salience Sentence Patterns in Automatic Abstracting Systems

自动文摘系统中一个关键的问题是找出能构成摘要的重点句子。找出这些句子的方法很多，但用机器学习的方法却较少，该文提出了一种关于文摘句式的自动学习方法。该方法以经过简单的预处理的若干语句为训练样本集，以正例句为基点进行由底向上的泛化学习，抽象出关于句式的一般概念，形成句式规则集，作为判断文中哪些语句可作为文摘句的有效手段。这是文摘系统实现的核心部分。

A key problem in automatic abstracting is to find salience sentences which can be included in the summary. There are many methods to get these sentences, but few with machine learning. This paper describes the use of machine learning on a training corpus of sentences to discover rules of salience sentences. An algorithm of sentence pattern learning is proposed,which generalizes those positive sentences from bottom to top. After training and learning from a corpus, a set of sentence rules regarding abstract is set up and will play a very important role in systems Of automatic abstracting.