一种网络聊天机器人的研究与实现

发布时间：2018-03-02 04:07

本文关键词： 聊天机器人服务匹配 REST 分词微服务　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着语音识别和自然语言处理技术的进步,聊天机器人(Chatbot)以其更友好体验被看成是人机交互的未来,市场上出现了大量的聊天机器人产品。然而,当前基于知识检索模型或生成模型的聊天机器人远没有达到期望的效果,甚至不实用。伴随着当前Web服务/APIs的大爆发,在这一背景下,本文提出了一种基于服务匹配的聊天机器人。该系统根据用户的输入,借助中文自然语言处理技术进行服务匹配,通过调用Web服务来响应用户意图。该系统可以解决知识检索模型维护知识库困难的问题以及生成模型很不成熟的问题,能给用户返回直接的和结构化的信息,在一定程度上使得聊天机器人可以代替现有APP。本文提出了聊天机器人服务匹配算法。通过中文分词、词性标注和命名实体识别等技术从用户输入的文本中提取关键词和命名实体,然后通过关键词搜索合适的API,根据API的信息,将提取的关键词和命名实体与API的调用参数匹配。本文采用CRF-BiLSTM模型进行中文分词,基于SIGHANBakeoff2005数据集对该模型分词效果进行了实验分析。结果证明较基于词典的机械分词方法MWSEG,CRF-BiLSTM的准确率、召回率和F值有接近10%的提升,达到了 Bakeoff 2005大赛Open组的最好水平。本文提出了适用于网络聊天机器人的RESTbot风格。RESTbot基于REST风格实现,针对网络聊天机器人的使用场景做了进一步的规范。聊天机器人调用RESTbot风格的Web服务无需人工编写格式转换代码,直接供前端使用,减少了重复的无意义的数据格式转换工作。然后本文针对RESTbot风格,提出了推荐的可视化方案。最后对RESTbot的可扩展性进行论述。最后,本文采用微服务架构实现了该网络聊天机器人系统。本文对网络聊天机器人系统的微服务架构、API网关、服务管理器和Web服务的实现细节进行了介绍。
[Abstract]:In recent years, with the development of speech recognition and natural language processing technology, chat robot Chatbott is regarded as the future of human-computer interaction with its more friendly experience. However, there are a lot of chat robot products on the market. The current chat robot based on knowledge retrieval model or generation model is far from achieving the desired effect or even being practical. In this context, with the explosion of current Web services / APIs, In this paper, a chat robot based on service matching is proposed. The system can solve the problem that the knowledge retrieval model is difficult to maintain the knowledge base and generate the immature model, and can return the direct and structured information to the user, by calling the Web service to respond to the user's intention, the system can solve the problem that the knowledge retrieval model is difficult to maintain the knowledge base. To a certain extent, the chat robot can replace the existing app. This paper proposes a matching algorithm for the chat robot service. Parts of speech tagging and named entity recognition are used to extract keywords and named entities from the text input by the user, and then search the appropriate API through keywords, according to the information of API. The extracted keywords and named entities are matched with the call parameters of API. This paper uses the CRF-BiLSTM model for Chinese word segmentation. The experimental results based on the SIGHANBakeoff2005 data set show that the accuracy of the model is higher than that of the dictionary-based mechanical word segmentation method MWSEG / CRF-BiLSTM, and the recall rate and F value are improved by nearly 10%. It reaches the best level of Open group in Bakeoff 2005 contest. This paper presents the implementation of RESTbot style. RESTbot is based on REST style, which is suitable for network chat robot. For the use of network chat robot to do a further specification. Chat robot calls RESTbot style of Web services without manual writing format conversion code, directly for the front-end use, This paper proposes a recommended visualization scheme for RESTbot style. Finally, the extensibility of RESTbot is discussed. In this paper, the microservice architecture is used to implement the web chat robot system, and the implementation details of the micro service architecture such as API gateway, service manager and Web service are introduced in this paper.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】