【作者】 梅洁；

【导师】 朱欣焰； 呙维；

【作者基本信息】 武汉大学 ， 地图学与地理信息系统， 2019， 硕士

【摘要】 随着室内外定位技术的迅速发展和移动存储设备处理能力的不断提高,基于位置的服务越来越多地出现在人们的日常生活中,人们通过移动终端即可获取所需要的与位置相关的各种生活服务信息。在检索信息时,人们往往倾向于选择关键词或者自然语句问句的形式表达自己的需求,这样便要求检索服务能够快速准确地对用户所输入问句进行语义理解。用户所输入的问句长短不一,形式多样,多含口语化表述习惯,难以使用规则或模板进行归纳。针对如何理解用户所输入问句的实际查询需求并检索得到关联信息这一问题,本文提出一种基于主题归类的位置信息问答式检索方法。对用户所输入问句进行预处理后将其归类到对应的检索主题,根据该主题的检索策略抽取问句中的有效信息,对位置本体数据进行结构化语义检索获得符合条件的实体结果集,基于检索结果信息进行图文展示,实现对问句的即时响应。首先,分析位置信息的组成内容并在此基础上定义位置本体的逻辑结构,实现语义位置信息本体建模存储,为问答式检索服务提供数据基础。其次,介绍基于样例和语义序列的主题归类方法,将待处理问句与主题样例库中问句进行比较,根据N-Gram模型计算其语义序列而非原始文本的相似度,匹配最相似样例的主题作为结果,实现样例量少而准确率高的效果,并选用Elasticsearch作为工具对主题样例库构建和样例最邻近匹配进行具体实现。然后,分析本体查询的特点,将问句看作(检索条件)概念(查询项)的组合,根据自定义的主题检索策略,自动抽取问句中关于检索条件和查询项的有效信息,生成问句查询语义图并构造SPARQL查询语句以执行语义检索操作,避免对复杂问句进行文法分析。最后,选取商场购物为应用背景,对问答式检索方法进行具体实践。分析室内商场数据特点,进行位置本体建模,结合已有商场服务研究和用户调研的方式总结商场问答式检索需求,整理主题词库,收集自然语言问句样例,划分主题并设计各主题的检索策略,构建商场问答式检索应用后对问答结果进行实验与分析。更多还原

【Abstract】 With the rapid development of indoor and outdoor positioning technology and the continuous improvement of mobile storage device processing capacity,location-based services are increasingly appearing in people’s daily life.People can easily obtain various location-related life service information they need through mobile terminals.In information retrieval,people tend to choose keywords or natural sentence questions to express their query intention,which requires the retrieval service to be able to quickly and accurately understand the semantic meaning of the user input questions.The natural language questions entered by users vary in length and form,and often contain colloquial expressions,making it difficult to use rules or templates for induction.Aiming at the problem of how to understand the actual query requirements of the user input and retrieve the relevant information,this paper proposes a question and answer retrieval method of location information based on topic classification.After preprocessing the questions entered by users and classifying them into corresponding retrieval topics,extract the valid information in the question according to the retrieval strategy of the topic,then carry out structured semantic retrieval of the location ontology data to obtain the specified entity result set,and based on the information of the retrieval results,the content combining graph and text are displayed to realize the instant response to the question.Firstly,define the logical structure of the location ontology based on the analysis of the connotation of the location information,so as to realize the modeling and storage of semantic location information based on ontology and provide data basis for questionand-answer retrieval service.Secondly,this paper introduces the method of topic categorization based on sample and semantic sequence in detail.This method calculates the similarity between the semantic sequence of the current question and the questions in the sample library rather than the original text according to the N-Gram model,and then gets the topic of the most similar sample to be matched as the result,achieving high accuracy based on a small number of samples.This paper selects the Elasticsearch as a tool to implement the construction of sample library and proximity matching between the current question and samples.Then this paper analyzes the characteristics of ontology query and regards the question entered by user as a combination of retrieval condition set,query item set and concept.According to the custom topic retrieval strategy,extract the effective information about the retrieval condition and query item in the question automatically,generate the query semantic graph and construct the SPARQL query statement to perform the semantic retrieval operation,avoiding the grammar analysis of the complex question.Finally,this paper select the shopping of mall as the application background,and put the question-and-answer retrieval method into practice.First of all,establish the location ontology model based on the characteristics of indoor shopping mall data,secondly sort out the basic lexicon and collect natural language questions after summarizing the question-and-answer retrieval requirements of shopping malls by combining the existing shopping mall service research and user survey,and then divide different topics and design retrieval strategies for each topic.After the application of question and answer retrieval in shopping mall is completed,test and analyze the result.更多还原