Web Intelligence Functions based on Text Processing

万维网（或简称万维网）已经发展成为一种新的全球信息流动和共享基础设施。为了增强网络对创造性个人、组织和社会的功能，我们从文本处理的角度对网络智能功能进行了研究。以下是主要成就。1)多文档摘要中重要句子提取的排序新方法：该方法不仅考虑了句子的时间顺序，还考虑了原文的先后关系和主题密切性。这种方法在国际上具有很高的新颖性。2)从网络中挖掘实体之间的关系：遵循人类（特别是研究人员）的关系挖掘系统，我们开发了一种从网络中挖掘公司之间关系的方法和系统。此外，我们还研究和开发了一种通用框架，用于挖掘网络中各种实体之间的关系。3)…更多文本情感感知：作为文本意义理解的一种，我们已经开发了两种从文本表达中提取情感或情感的方法。一种方法包含了使用常识语料库的高级功能。本文还提出了一种基于情感亲和力的网络新闻分类系统。4)基于Web搜索引擎的词间语义相似度计算方法：首先，我们开发了一种基于Web搜索引擎的AND搜索命中次数计算词间语义相似度的方法。然后，为了提高性能，我们设计了一种有效的方法，该方法利用搜索引擎输出的片段信息，并考虑目标两个单词的局部模式。5)自然语言文本的通用概念描述语言：作为面向下一代Web的基础，我们开展了建立通用概念描述语言CDL.nl的研究。与旨在描述元数据的语义网不同，CDL。自然语言允许以机器可理解的形式描述自然语言文本中隐含的概念含义。这项工作是与我们的cdl - 1国际标准化活动同时进行的。少

The World Wide Web (or just Web) has grown to a new global infrastructure for information flow and sharing. In order to enhance Web functions toward creative personals, organization and societies, we have conducted our research on Web intelligence functions especially from the viewpoint of text processing. Below are the main achievements.1) A new ordering method of extracted important sentences in multi-document summarization: This new method considers not only chronological ordering of the sentences, but also precedence and succession relations in the original texts and topical closeness. This approach possesses high novelty in international level.2) Relation mining between entities from the Web: Following a human (especially, researcher) relation mining system, we have developed a method and a system for mining relations between firms from the Web. Also, we have undertaken a research and development on a generic framework that mines relations between various entities from the Web.3) … More Textual affect sensing: As one type of text meaning understanding, we have developed two methods that extract affect or emotion from textual expressions. One method has incorporated an advanced function of employing a corpus of commonsense. A classification system of Web online news articles based on their emotion affinity has also been developed.4) Method for computing semantic similarity between words using a Web search engine: First, we developed a method for computing similarity between words based on the hit counts of the AND search of a Web search engine. Then, in order to improve the performance, we devised an effective method that also employs snippet information of the search engine's output and takes local patterns with the target two words into account.5) A Common Concept Description Language for natural language texts: As a basis toward next-generation Web, we have carried out our research to set up a common concept description language called CDL.nl. Unlike the Semantic Web which aims to describe meta-data, CDL.nl allows to describe the concept meaning implied in natural language texts in a machine understandable form. This work has been proceeded in parallel with our international standardization activity of CDL.nl. Less