【作者】 王俊；

【导师】 饶泓； 赵聪；

【作者基本信息】 南昌大学 ， 计算机技术（专业学位）， 2022， 硕士

【摘要】 随着人们生活水平的提高,人们越来越重视健康问题,健康饮食成为人们关注的热点,目前人们大多通过科普文章或短视频方式片段获取饮食健康知识,往往不能直接获得答案,需要筛选有用信息。因此,构建一个有效且实用的智能饮食健康问答系统成为急需解决的问题。针对这个问题,本文收集并处理海量饮食健康数据,构建营养健康知识图谱,基于知识图谱和自然语言处理技术设计与实现了一个营养健康智能问答系统,以解决人们日常生活中的饮食健康的知识获取问题。本文主要工作如下:(1)构建饮食健康领域知识图谱。目前开源的中文知识图谱还没有饮食健康领域相关的知识图谱,本文通过Scrapy爬虫技术爬取饮食健康相关的网站数据,再与已有项目数据库数据结合来构建知识图谱的源数据,本文利用Protégé本体建模软件设计本体,利用构建的本体和领域知识构建饮食健康领域知识图谱。(2)研究并实现了基于领域命名实体识别的数据集自动化标注技术。通过构建领域实体词典,基于词典的双向最大匹配算法为待标注句子进行分词,再通过脚本对句子进行B-I-E-S-O标注。(3)以本文构建的饮食健康知识图谱为数据支撑研究饮食健康知识智能问答任务。本文将问答算法分为问句属性链接和命名实体识别这两个任务,首先对用户提出的问句使用基于BERT-BiLSTM-CRF命名实体识别模型,将问句中的领域实体标识出来,同时构造出该问句对应的候选属性集,再使用基于BERT的问句属性链接模型确定问句对应实体属性,基于关键特征和规则,在图数据库Neo4j中查找答案。(4)设计并实现了饮食健康知识问答系统,系统采用Spring、SpringBoot、SpringMVC、MyBatis架构。模型端通过Flask架构做成API接口的形式供系统调用,系统功能模块分为命名实体识别功能模块、水果蔬菜展现模块、实体详情模块、智能问答模块四个模块。智能问答为其中的核心模块,能够智能回答用户提出的饮食健康领域的问题。更多还原

【Abstract】 With the improvement of people’s living standards,people pay more and more attention to health issues,and healthy eating has become a hot spot of people’s attention.At present,most people acquire knowledge about diet and health through popular science articles or short video clips,but often cannot get the answer directly,and need to filter useful information.Therefore,building an effective and practical smart diet health question answering system has become an urgent problem to be solved.In response to this problem,this paper collects and processes massive diet and health data,constructs a nutrition and health knowledge graph,and designs and implements a nutrition and health intelligent question answering system based on knowledge graph and natural language processing technology to solve the knowledge acquisition of dietary health in people’s daily life.question.The main work of this paper is as follows:(1)Build a knowledge graph in the field of diet and health.At present,the opensource Chinese knowledge graph does not have a knowledge graph related to the field of diet and health.This article uses the Scrapy crawler technology to crawl the website data related to diet and health,and then combines with the existing project database data to construct the source data of the knowledge graph.This article uses the Protégéontology.The modeling software designs ontology,and uses the constructed ontology and domain knowledge to build a knowledge map in the field of diet and health.(2)Research and implement the automatic labeling technology of datasets based on domain named entity recognition.By constructing a domain entity dictionary,the dictionary-based bidirectional maximum matching algorithm performs word segmentation for the sentences to be marked,and then performs B-I-E-S-O marking on the sentences through scripts.(3)Using the dietary health knowledge map constructed in this paper as the data support to study the intelligent question answering task of dietary health knowledge.In this paper,the question answering algorithm is divided into two tasks: question attribute linking and named entity recognition.First,the BERT-BiLSTM-CRF-based named entity recognition model is used for the question proposed by the user,and the domain entities in the question are identified and constructed at the same time.Obtain the candidate attribute set corresponding to the question,and then use the BERT-based question attribute link model to determine the corresponding entity attribute of the question,and find the answer in the graph database Neo4 j based on key features and rules.(4)Design and implement a question and answering system for dietary health knowledge.The system adopts Spring,SpringBoot,SpringMVC,and MyBatis architecture.The model end is made into the form of an API interface through the Flask architecture for system calls.The system function modules are divided into four modules: named entity recognition function module,fruit and vegetable display module,entity details module,and intelligent question answering module.Intelligent question answering is the core module,which can intelligently answer the questions raised by users in the field of diet and health.更多还原