Concept Chemical Knowledge Representation

近年来，社会上有大量以自然语言文本为主的电子文档通过万维网等进行存储和传播。虽然它们包含了各种有用的知识，但是对于我们和计算机来说，很难将它们作为知识来充分利用，比如将多个知识组合起来得出一个答案。为了解决这一问题，我们提出并开发了一种知识表示/推理方案，称为“概念化学表示（CCR）”，该方案的表示接近自然语言表达，从而便于从自然语言句子转换。在将自然语言文本高效地转换为CCR时，需要关注有意义的句子，忽略其他部分，避免将无用的成分纳入CCR知识库。因此，我们开发了以下从文档中提取关键字和重要句子的方法。1)利用词共现偏差统计进行关键词提取。2)利用词共现的小世界结构提取关键词。3)基于词汇活动的关键词提取，作为人类词汇识别的认知过程。现有的大多数方法都是基于TF*IDF（词频*逆文档频率）的使用，而我们的上述原始方法与现有的方法有所不同。我们的方法既适用于英语句子，也适用于日语句子。

A huge quantity of electric documents mostly consisting of natural language texts are stored and circulated recently through the World Wide Web (WWW), etc. in the society. Although they include a variety of useful knowledge, it has been difficult for us and computers to fully utilize them as knowledge, such as deriving an answer by combining multiple pieces of knowledge. Toward one solution to this problem, we have proposed and developed a knowledge representation/inference scheme called "Concept Chemical Representation (CCR)", which representation is close to natural language expression and thus convenient for the conversion from natural language sentences.When converting natural language texts into CCR efficiently, it is required to focus meaningful sentences and ignore other parts for avoiding the inclusion of useless components into the CCR knowledge base. Accordingly, we have developed the following extraction methods of keywords and important sentences from a document.1) keyword extraction using the deviation statistics of word co-occurrence.2) keyword extraction using the small world structure of word co-occurrence.3) keyword extraction based on term activities measured as the human cognitive process for term recognition.While most existing methods are based on the use of TF*IDF (term frequency * inverse document frequency), our above original methods are different from the existing ones. Our methods can be applied to English sentences as well as Japanese sentences.