基于领域本体的客服问答系统的设计与实现

发布时间：2019-07-01 10:40

【摘要】：目前国内电子商务平台的企业和商家主要依赖客服人员来解决用户售前售后的提问。为减少客服的工作量,提高系统的效率,本文将用户常见的问题和领域知识结合构建知识库,实现客服问答系统。目前领域本体在问答系统中的应用主要体现在对领域概念的分类方面,缺乏对用户问句的语义理解和处理。本文以母婴领域为例,本文分析客服系统的聊天记录,构建领域本体和FAQ库,并重点研究和改进了领域本体对实体的属性描述和属性相似度计算方法,实现了基于领域本体的客服自动问答系统。在以下方面展开了研究工作:首先研究了领域本体概念及其关系的表达,构建了母婴知识的领域本体。在本体的构建过程中,本文不仅对领域的概念分类关系进行表示,而且强调对实体属性的描述。利用句子依存语法分析将自然语言的主谓宾关系与本体中的主谓宾关系进行映射。其次,研究了领域本体的概念相似度计算方法,对基于本体的概念相似度算法提出改进。以往计算本体的概念相似度所提到的“概念”对应本体中的类或实体,本文根据构建的领域本体,设计并实现了本体中的类、实体和属性的相似度计算方法,实现了概念之间的非分类关系的语义相似度计算。最后,根据改进的概念相似度计算方法,提出了一种句子的本体扩展方法。该方法根据类的分类关系判断删除句子的冗余信息,并且根据类的非分类关系对句子的潜在信息进行扩展。最后结合VSM算法,进行句子的相似度计算,建立了可对母婴商品和相关知识查询的问答系统。目前该算法模型已成功应用于青岛市海尔软件智能客服问答系统HelloOWN中。使用母婴本体后,该系统在语义分析方面取得了良好效果。
[Abstract]:At present, domestic e-commerce platform enterprises and businesses mainly rely on customer service personnel to solve users' pre-sale and post-sale questions. In order to reduce the workload of customer service and improve the efficiency of the system, this paper combines the common questions and domain knowledge of users to build a knowledge base to realize the customer service question and answer system. At present, the application of domain ontology in question answering system is mainly reflected in the classification of domain concepts, and lacks the semantic understanding and processing of user questions. Taking the mother and child domain as an example, this paper analyzes the chat records of the customer service system, constructs the domain ontology and FAQ library, and focuses on the research and improvement of the attribute description and attribute similarity calculation method of the domain ontology to the entity, and realizes the customer service automatic question and answer system based on the domain ontology. The research work has been carried out in the following aspects: firstly, the concept of domain ontology and the expression of its relationship are studied, and the domain ontology of maternal and infant knowledge is constructed. In the process of ontology construction, this paper not only represents the conceptual classification relationship of domain, but also emphasizes the description of entity attribute. Sentence dependent grammar analysis is used to map the subject-predicate-object relationship of natural language with the subject-predicate-object relationship in ontology. Secondly, the concept similarity calculation method of domain ontology is studied, and the ontology-based concept similarity algorithm is improved. In the past, the "concept" mentioned in calculating the concept similarity of ontology corresponds to the class or entity in ontology. According to the constructed domain ontology, this paper designs and implements the similarity calculation method of class, entity and attribute in ontology, and realizes the semantic similarity calculation of non-classification relationship between concepts. Finally, according to the improved concept similarity calculation method, a sentence ontology extension method is proposed. This method judges the redundant information of deleted sentences according to the classification relationship of the class, and extends the potential information of the sentence according to the non-classification relationship of the class. Finally, combined with VSM algorithm, the similarity of sentences is calculated, and a question and answer system which can query maternal and infant goods and related knowledge is established. At present, the algorithm model has been successfully applied to Qingdao Haier software intelligent customer service question answering system HelloOWN. After using maternal and infant ontology, the system has achieved good results in semantic analysis.
【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【相似文献】