典型事件过程建模的研究

项目名称： 典型事件过程建模的研究

项目编号： No.60803094

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 李晗静

作者单位： 哈尔滨工业大学

项目金额： 20万元

中文摘要： 典型事件过程是指符合常识的，此事件从开始到结束中涉及的一系列子事件及它们之间的过程关系。目前和事件相关的研究主要集中于事件识别、事件时序等一些关系抽取，还没有在文献中发现对典型事件过程建模的研究。而这类知识可以用于NLP领域和文景转换中与事件推理有关的各种应用中。本课题提出了基于常识语料库的典型事件过程建模方法，也就是形式化为有向事件图的方法。课题以英语为研究对象，从普遍意义角度，设计表示典型事件过程的事件图；提出用相对熵表示事件之间有向的关联强度，从而确定子事件；研究基于数据挖掘、信息抽取方法的事件图生成技术；以wiki网站内容和儿童故事文本为资源，构建常识丰富的语料库。由于传统金标准（golden-standard）评价的昂贵代价，提出基于实验和应用的的双层评价体系。前者负责评测各种特征对该方法性能的影响；后者在动画自动生成任务上检查典型事件过程模型的有效性。

中文关键词： 典型事件过程；事件图；常识获取；文景转换

英文摘要： Typical event procedure is the temporal relationships between a series of sub-events which are composed of the event from its beginning to ending timely. The art of the state of researches on events include event identity, event temporal retrieval and some like this. On the best of our knowledge, there are few paper talked on the term of the typcial event procedure. The knowledge of the typical event procedure can be applied in NLP, text-to-scene conversion, and event inference. The program proposes the model of the typical event procedure on common corpus, that is to be as event graph formally. The aim of the program is to design event graph for the typical event procedure on the subject of English. The method of spliting event into a series of sub-event is the directing relevant degree on relative entropy. The generation technoloies on data mining and information retrieval are talked deeply.The common sense corpus is built on wiki and stories. The program propsals two levels evaluation standard in that the golden-standard is expensive. The first level charges of every method efficience on features, and the second is for the model of the typical event procedure on the task of animation automation generation.

英文关键词： The typical event procedure; event graph;common sense capture; text-to-scence conversion