【作者】 王冲；

【导师】 俞勇；

【作者基本信息】 上海交通大学 ， 计算机应用技术， 2008， 硕士

【摘要】 语义搜索是语义网研究领域的一个重要内容。通过充分利用底层数据丰富的语义信息,语义搜索力图为用户直接提供精准的答案,而不仅是包含有所查关键字的文档。然而,这种语义搜索需要用户输入复杂的形式化查询语句,并需要用户对底层数据的模式和知识库词汇有一定了解,而这样的要求是普通用户所不能达到的。因此,如何提供一种易于让广大普通用户使用的友好的语义搜索用户接口是一项有意义的工作。本文立足于现有的语义网技术,对本体及其他结构化数据进行了分析,提出了一种支持用户通过输入自然语言或关键字来表达信息需求的语义搜索接口框架。本框架中,通过利用本体知识和自然语言处理技术及工具,用户查询被匹配到本体中的对象,然后转换为三元组中间结构,最后被翻译为可被现有形式化查询引擎优化执行的SPARQL语句。在从自然语言查询到形式化查询的转换过程中我们研究了自然语言查询语法解析和语义转换两个主要困难并有针对性提出了两阶段的转换方式;在语法信息利用上,我们提出了基于基本名词短语的分析方法。在关键字查询的转换过程中,我们重点考虑了歧义性的问题,并提出了对转换结果进行排序的方法。实验表明,本文提出的方法在转化查询的质量上达到了比较好的效果,也证明通过自然语言和关键字可以进行有效的语义搜索,是有吸引力的语义搜索用户接口形式。此外,我们还分析了目前方法的不足之处,并对进一步的研究进行了讨论。更多还原

【Abstract】 Semantic search is one important research topic in the Semantic Web community. By making full use of the semantics embedded in the underlying data, semantic search promises to provide the users with precise answers, rather than documents which only contain the typed keywords. However, such semantic search requires the users to input formal logic queries, as well as some knowledge about the schema and vocabulary of the underlying knowledge base, which is beyond the desire and capability of casual users. Therefore, it is valuable to study on how to provide the casual users a friendly semantic search interface.In this paper, by making use of the current Semantic Web technologies and studying on other similar structural data, we propose a framework which accepts natural language or keyword queries for semantic search. Within this framework, by utilizing ontology knowledge and existing natural language processing techniques and tools, the user input queries are first mapped to entities in the ontology and then transformed to an intermediate data model called triples, which is finally interpreted as formal SPARQL queries that can be optimized and executed by a state-of-the-art engine. During the process of translating a natural language query into formal logic query, we analyze the major difficulties and propose a two-stage transformation approach accordingly. We also propose a BaseNP-based analyzing technique to extract syntactic information from a parse tree. While translating a keyword query, we pay most of our efforts on disambiguating the multiple results produced, and propose a ranking method for that.The experiments show that the proposed approach achieved a good performance on translation quality, which also prove that natural language and keyword interface are effective approach for semantic search. In addition, we also analyze the limitations of our approach and discuss on future work.更多还原