全信息理论在自动文摘系统中的应用

The Application of Comprehensive Information Theory in Automatic Abstract System

该文介绍了全信息理论提出的背景及其主要内容，并将其应用到一种智能业务──自动文摘系统中，设计实现了一个基于理解的、面向神经网络学习算法领域的中文９动文摘系统 Ｌａｄｉｅｓ。该系统综合利用语法信息和语义信息来理解原文的句法和含义，并通过语用信息制导，对原文进行文摘信息的过滤、提取和生成高质量的文摘。实验结果证明，这一方法是行之有效的。

We introduce the background and major contents of the comprehensive information theory and apply it to the research of automatic abstract in this paper. We have designed and implemented a Chinese automatic abstract system named Ladies (Literature Abstract-and-Digest information Extract System), which is special for texts of the neural networks' algorithms field. Using the syntactic and semantic information to understand the text's structure and meaning, and pragmatic information to drive, Ladies can filter the text's information and extract and generate abstracts. The results of experiments have shown that the whole set of methods is highly effective.