【作者】 仲兆满；

【导师】 刘宗田；

【作者基本信息】 上海大学 ， 计算机应用技术， 2011， 博士

【摘要】 本体通过明确地定义概念和概念间的关系描述事物或现象的本质。本体已经成为人工智能和知识工程中一种先进的技术,在知识的表示、获取和应用等方面发挥着越来越重要的作用。国内外许多研究机构纷纷对其展开了广泛的理论研究和应用探索。近来,“事件”的概念逐渐被计算语言学、人工智能、信息检索、信息抽取、自动文摘和自然语言处理等知识处理领域所采用。大量的文本,比如小说、史传、回忆录、神话传说、民间故事、叙事诗、戏剧、人物传记、新闻报道等,都包含有各类事件。“事件”关联了参与者、时间和地点等概念,是比“概念”粒度更大的知识单元。人类遭受过并且正在遭受着各种灾难性突发事件的危害,包括地震、火山爆发、洪水、飓风、化学品泄漏、核辐射逃逸、传染病、事故、爆炸、城市火灾等等。由于现实中的事件在网络上都有明显的反映,借助搜索引擎从互联网上获取事件相关信息已经是用户的迫切需求。但由于互联网上的信息急剧膨胀,通用搜索引擎返回的结果往往是信息量大、查询不准确。用户在输入某个关键字后,搜索到的有用信息并不多,对事件类信息的检索需求更是如此。因此,面向事件的信息检索技术亟待研究开发。本文围绕事件本体及其在查询扩展中的应用,研究了事件类之间的关系、事件类关联强度的量化,在此基础上提出了一种面向事件的本体模型。接着,将事件的思想应用于查询扩展领域,探索了基于伪相关反馈和基于事件本体的面向事件的查询扩展方法。本文的研究内容和创新点主要包括:(1)定义事件类,依据事件类的动作要素,分析事件类之间存在的关系;依据事件类的参与者、时间、环境等要素,揭示事件实例存在的关系。定义事件类影响因子的概念,用来描述一个事件类实例的发生对有关系的其他事件类的实例所产生的影响大小。在事件类、事件类关系、事件类影响因子的基础上,改进并完善了事件本体模型。(2)选取突发事件领域的一个子领域研究事件本体的构建方法。分析动态事件与静态概念在相互关联上的区别,根据事件具有的随时间而动态变迁的特性,提出了综合考虑事件的Authorities值和Hubs值的事件重要度的计算方法(记作HARank),并将此算法应用于文本集合中重要事件的识别、事件本体中事件类重要度的计算。(3)针对用户获取互联网上事件类信息的需求,提出面向事件的查询扩展技术。方法之一是面向事件的基于伪相关反馈查询扩展方法,重点研究文本中事件的识别、查询项中限定项与事件项的判别、扩展事件的选取、查询项的权值设置以及查询项与文档相似度的计算等问题;方法之二是基于事件本体面向事件的查询扩展方法,重点研究基于事件本体的查询项中事件项的识别、查询项中事件项到事件本体的事件类、事件类到其各个要素的联想扩展。更多还原

【Abstract】 Ontology reflects the essence of objects or phenomenon through defining concepts and their relations explicitly. Ontology has become an advanced technology for artificial intelligence and knowledge engineering, which plays an increasingly important role in the representation, acquisition and application of knowledge. Therefore, many research institutions at home and abroad have launched extensively theoretical research and application exploration.Recently, the‘event’has been applied to computational linguistics, artificial intelligence, information retrieval, information extraction, automatic summarization and natural language processing. The various texts such as novel, biography, memoir, myth and legend, folktale, epic, drama, press, and so on, include lots of events. An event identified by event triggers is associated with participants, time, location, etc, which is a larger knowledge unit than a concept.The human have suffered and have been suffering all kinds of emergency, including earthquakes, volcanic eruptions, floods, hurricanes, chemical spills, nuclear radiation escape, diseases, accidents, explosions, urban fires, etc. Because an important event in the real world is always reflected on the network in different styles, getting event information has become the key component for the users. With an overwhelming volume of information currently available, the results returned by universal search engine are often informative and inaccurate, and users’experiences of searching are not very good. It is more so for the retrieval of events. So it is urgent to study and develop the event-oriented retrieval technology.The work surrounds event ontology and its application in query expansion in this paper. First, we study the relations of event classes and the association of event classes, and propose an event-oriented ontology model. Then, event is applied in query expansion area, and we explore the methods of query expansion based on pseudo relevant feedback and event ontology. The main contents and innovations of this paper include:(1) Define the concept of event class, analyze the relations of event classes according to the action element of the event class and the relations of event instances according to the object, time and environment elements of the event class. Define the influence factor between event classes to depict the probability by that if an event instance occurred, the other event instance occurs too. Improve an event-oriented ontology model on the basis of the event class, the event class relation and the event class influence factor.(2) Select a sub-area of emergency to study the method of constructing event ontology. Analyze the distinctions of dynamic event and static concept. Put forward a method of computing the importance of events, which synthetically considers both the hubs and authorities of events, denoted by HARank (Hubs-Authorities Rank), and apply the HARank to identify important events from the collection of texts and rank event classes for event ontology.(3) Propose the technology of event-oriented query expansion aiming at the requirements of getting event information. First, present the method of event-oriented query expansion based on pseudo relevant feedback, including the identification of events from texts, the discrimination of qualified terms and event terms of query terms, the selection of expansion events, the setting of query term weights and the similarity computation between query terms and texts. Second, present the method of event-oriented query expansion based on event ontology, including the identification of event terms of query terms based on event ontology, and the associative expansions from event terms of query terms to event classes of event ontology and from the event class to its elements.更多还原