【作者】 张岩；

【导师】 尹铁岩；

【作者基本信息】 吉林大学 ， 管理科学与工程， 2018， 硕士

【摘要】 近年来互联网快速发展,传统的线下交易已经在互联网的带动下大量转移到了线上。线上交易的虚拟性导致它存在着商品质量难以保障,用户咨询困难等不足,除此之外,消费者越来越关注服务的质量,在线上交易的过程中客服的参与十分重要。但目前我国的电子商务市场规模大,网购用户数量极多,客服人员面临着极大的工作压力,此外,客服还有着招聘困难、人力成本高、流失率居高等特点,这样的困境不利于我国电子商务的良性发展,亟待解决。为了解决这些问题,应用相关技术实现客服的自动应答是一种很好的手段,可以在解决问题的同时进一步提高客户服务的响应速度和服务质量,降低企业的人力成本。本文对问答系统的结构及已有研究成果进行了梳理,从问答系统的定位、功能与相关技术进行了简述,完成了相关的知识储备,进而分析了智能应答中存在的几个重要环节,即问句理解,信息检索,答案抽取及排序等。问答系统允许用户使用自然语言,在理解用户的问题后,向用户提供对其问题直接而准确的答案。根据应用领域进行分类,问答系统可分为开放领域问答系统和封闭领域问答系统,而智能客服系统则是问答系统在封闭领域中的一个应用,而结合用户场景对于提升智能客服系统的准确率十分有效,本文在此基础上,进行了智能客服系统的研究与设计。本文首先收集了大量的客服咨询语料及用户所处状态,在分析筛选后结合常见的停用词构建相应的停用词表,并使用哈工大LTP分词技术进行处理,构建了手机相关的专业词典来提高分词的准确度,在现有分词的基础上,进行词类标注、同义词替换等操作,然后根据咨询语料完成FAQ库的构建,结合用户场景优化答案的配置。之后提取用户输入内容中的关键词,使用word2vec找关键词的相似词,检测是否存在业务关键词进行过滤,如果不存在则识别为闲聊,转到已经比较成熟的图灵机器人API,如果存在的话则转到搭建的智能客服系统。然后结合用户所处场景,根据用户输入的内容与FAQ库中的问题进行匹配,设定相似度的阈值并进行调优,若相似度足够高,则直接返回问题的问题答案对;若相似度高于设定的阈值,则根据关键词、模糊匹配等返回相似度较高的问题列表,由用户选择后返回相应的答案;否则匹配失败,转到人工客服进行处理。最后,使用历史咨询语料对系统的答疑能力进行验证,通过测试可以发现,本文的智能客服系统可以在一定程度上满足消费者的咨询需求。更多还原

【Abstract】 In recent years,the Internet has developed rapidly.The traditional offline transactions have been largely transferred to the Internet driven by the Internet.Due to the virtual nature of online transactions,it is difficult to ensure the quality of the products and the difficulty of user consultation.In addition,consumers are increasingly concerned about the quality of services.It is very important for the customer to participate in online transactions.However,at present,China’s e-commerce market is huge,and the number of online shopping users is very large.The customer service is faced with a great deal of work pressure.In addition,customer service also has difficulties in recruiting,high labor costs,and a high turnover rate.It is beneficial for the sound development of China’s e-commerce to be solved.In order to solve these problems,it is a good method to apply the related technology to realize the automatic response of the customer service,which can further improve the response speed and service quality of the customer service while reducing the human cost of the enterprise.This paper combs the structure of the question answering system and the existing research results,briefly describes the positioning,function and related technologies of the question answering system,completes the relevant knowledge reserves,and then analyzes several important links in the intelligent response.That is,question understanding,information retrieval,answer extraction,and sorting.According to the application areas,the question answering system can be divided into open field question answering system and closed field question answering system.The intelligent customer service system is an application of the question answering system in the closed field.Based on the requirements and related background of smart customer service system,this paper carries out research and design of smart customer service system.This article first collected a large number of customer service consulting corpus and the status of the user,after the analysis and filtering combined with common stop words to construct the corresponding stop word list,and use HIT LTP segmentation technology to deal with,build a mobile phone related professional dictionary Improve the accuracy of word segmentation,based on the existing word segmentation,word segmentation,synonym replacement and other operations,and then complete the construction of the FAQ database based on consulting corpus,and optimize the configuration of the answer based on user scenarios.After that,the keywords in the user input are extracted,word2 vec is used to find the similar words of the keywords,and it is detected whether there is a business keyword to filter,and if it does not exist,it is identified as chat,and it is transferred to the mature Turing robot API if there exists.If you do,go to the built smart customer service system.Then,according to the scene where the user is located,the content of the user’s input is matched with the questions in the FAQ database.The similarity threshold is set and the optimization is performed.If the similarity is high enough,the question and answer pair of the question is directly returned;if similarity Above the set threshold,a list of questions with a higher degree of similarity is returned according to keywords,fuzzy matches,etc.,and the user selects and returns a corresponding answer;otherwise,the match fails,and it is transferred to an artificial customer service for processing.Finally,using historical consulting corpus to verify the ability of the system to answer questions,through testing can be found that this article’s intelligent customer service system can meet the consumer’s consulting needs to a certain extent.更多还原