基于在线百科的文本蕴含知识获取.docxVIP

31 - 哈尔滨工业大学本科毕业设计（论文） 基于在线百科的文本蕴含知识获取 摘 要 由于自然语言在表达形式上灵活多样，相同的含义往往可以采用不同的陈述形式来表达，或从不同的文本中推断出来，这样的同义异形现象，使得文本处理变的十分困难。文本蕴含正是处理这种同义异形现象的有效途径。 文本蕴含，是两个文本间的有方向的关系，用T表示一个文本（Text，T），H表示另一个文本（Hypothesis，H），如果根据T的上下文语境和世界知识，可以从T的含义中推断出H的含义，则称T蕴含H。 为了识别文本蕴含，需要依赖外部的世界知识，知识获取应运而生。不过目前中文的文本蕴含知识资源较少，本文的目的就是获取中文上谓词间的语境敏感的蕴含关系。 我们采用的是两级语境敏感模型：基于词级别分布假设的模型结合主题模型的方法。基于分布假设的模型采用 DIRT,主题模型的构建采用 LDA。在英文以及中文的数据集上的评价显示，采用的两级语境敏感模型比之前单独的词级别和主题级别的模型有显著提高，并且在中文上得到一个语境敏感的谓词推断规则集。 关键词：文本蕴含；知识获取；语境敏感；两级模型；分布性假设DIRT；主题模型LDA Abstract The natural language expression in the form of flexible, the same meaning can often take different forms to express statement or be inferred from a different text, such Synonymy phenomenon, making text processing becomes very difficult. Textual Entailment is an effective way to deal with this precisely synonymous shaped phenomena. Textual Entailment, there is a relationship between the direction of the two texts, with T represents a text (Text, T), H represents another text (Hypothesis, H), according to context of T and knowledge of the world, from the T infer the meaning of H, then T entails H. In order to recognize the Textual Entailment, need to rely on the outside world knowledge, knowledge acquisition came into being. But now Chinese Textual Entailment knowledge with fewer resources, the purpose of this paper is to get context-sensitive on Chinese implied relationship between the predicate. We use a Two-Level Context Sensitive Model: the integration method based on word-level distributional hypothesis and topic-level context representations. Based on distributional hypothesis adopted DIRT, topic model constructed using LDA. Evaluation in the English and Chinese data sets show that the model used in a significant increase than the individual word-level and topic-level model before , and to get a context sensitive set of inference rules on Chinese . Keywords: Textual Entailment, Knowledge Acquisition, Context Senitive, Two-level Model, DIRT, LDA