SpringBoot集成图数据库neo4j实现简单的关联图谱

社交领域：Facebook, Twitter，Linkedin用它来管理社交关系，实现好友推荐

图数据库neo4j安装：

下载镜像：docker pull neo4j:3.5.0
运行容器：docker run -d -p 7474:7474 -p 7687:7687 --name neo4j-3.5.0 neo4j:3.5.0
停止容器：docker stop neo4j-3.5.0
启动容器：docker start neo4j-3.5.0
浏览器 http://localhost:7474/ 访问 neo4j 管理后台，初始账号/密码 neo4j/neo4j，会要求修改初始化密码，我们修改为 neo4j/123456

neo4j中可以使用Cypher查询语言（CQL）进行图形数据库的查询

①、添加节点

CREATE (:) 创建不含有属性节点

节点名称node-name和标签名称lable-name：标签名称相当于关系型数据库中的表名，而节点名称则代指这一条数据
而创建包含属性的节点时，可以在标签名称后面追加一个描绘属性的json字符串

CREATE (索尔:Person)CREATE (洛基:Person {name:"洛基",title:"诡计之神"})

②、查询节点

MATCH (:) 查询已存在的节点及属性的数据

MATCH命令在后面配合RETURN、DELETE等命令使用，执行具体的返回或删除等操作

MATCH (p:Person) RETURN p

在这里插入图片描述
可以看到上面添加的两个节点，分别是不包含属性的空节点和包含属性的节点，并且所有节点会有一个默认生成的id作为唯一标识

③、删除节点

MATCH (p:Person) WHERE id§=100
DELETE p

在这条删除语句中，额外使用了WHERE过滤条件，它与SQL中的WHERE非常相似，命令中通过节点的id进行了过滤。删除完成后，再次执行查询操作，可以看到只保留了洛基这一个节点

④、添加关联

再创建一个节点作为关系的两端：CREATE (p:Person {name:“索尔”,title:“雷神”})

创建关系的基本语法如下：

CREATE (:)

[:]
-> (:)

也可以利用已经存在的节点创建关系，下面我们借助MATCH先进行查询，再将结果进行关联，创建两个节点之间的关联关系：

MATCH (m:Person),(n:Person) 
WHERE m.name='索尔' and n.name='洛基' 
CREATE (m)-[r:BROTHER {relation:"无血缘兄弟"}]->(n)
RETURN r

添加完成后，可以通过关系查询符合条件的节点及关系：

MATCH (m:Person)-[re:BROTHER]->(n:Person) 
RETURN m,re,n

在这里插入图片描述
如果节点被添加了关联关系后，单纯删除节点的话会报错，：

Neo.ClientError.Schema.ConstraintValidationFailed
Cannot delete node<85>, because it still has relationships. To delete this node, you must first delete its relationships.

这时，需要在删除节点时同时删除关联关系：

MATCH (m:Person)-[r:BROTHER]->(n:Person) 
DELETE m,r

SpringBoot整合neo4j

一、依赖


<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd"><parent><artifactId>springboot-demo</artifactId><groupId>com.et</groupId><version>1.0-SNAPSHOT</version></parent><modelVersion>4.0.0</modelVersion><artifactId>neo4j</artifactId><properties><maven.compiler.source>8</maven.compiler.source><maven.compiler.target>8</maven.compiler.target></properties><dependencies><dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-web</artifactId></dependency><dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-autoconfigure</artifactId></dependency><dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-test</artifactId><scope>test</scope></dependency><dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-data-neo4j</artifactId></dependency><dependency><groupId>com.hankcs</groupId><artifactId>hanlp</artifactId><version>portable-1.2.4</version></dependency><dependency><groupId>edu.stanford.nlp</groupId><artifactId>stanford-parser</artifactId><version>3.3.1</version></dependency><dependency><groupId>org.projectlombok</groupId><artifactId>lombok</artifactId></dependency></dependencies>
</project>

二、属性文件和启动类


server:port: 8088
spring:data:neo4j:uri: bolt://127.0.0.1:7687username: neo4jpassword: 123456

三、文本SPO抽取

借助Git上一个现成的工具类，来进行文本的语义分析和SPO三元组的抽取工作
项目地址：https://github.com/hankcs/MainPartExtracto

//提取主谓宾
public class MainPartExtractor{private static final Logger LOG = LoggerFactory.getLogger(MainPartExtractor.class);private static LexicalizedParser lp;//加载模型private static GrammaticalStructureFactory gsf;static{//模型String models = "models/chineseFactored.ser";LOG.info("载入文法模型：" + models);lp = LexicalizedParser.loadModel(models);//汉语TreebankLanguagePack tlp = new ChineseTreebankLanguagePack();gsf = tlp.grammaticalStructureFactory();}//获取句子的主谓宾public static MainPart getMainPart(String sentence){// 去掉不可见字符sentence = sentence.replace("\\s+", "");// 分词，用空格隔开List<Word> wordList = seg(sentence);return getMainPart(wordList);}public static MainPart getMainPart(List<Word> words){MainPart mainPart = new MainPart();if (words == null || words.size() == 0) return mainPart;Tree tree = lp.apply(words);LOG.info("句法树:{}", tree.pennString());// 根据整个句子的语法类型来采用不同的策略提取主干switch (tree.firstChild().label().toString()){case "NP":// 名词短语，认为只有主语，将所有短NP拼起来作为主语即可mainPart = getNPPhraseMainPart(tree);break;default:GrammaticalStructure gs = gsf.newGrammaticalStructure(tree);Collection<TypedDependency> tdls = gs.typedDependenciesCCprocessed(true);LOG.info("依存关系：{}", tdls);TreeGraphNode rootNode = getRootNode(tdls);if (rootNode == null){return getNPPhraseMainPart(tree);}LOG.info("中心词语：", rootNode);mainPart = new MainPart(rootNode);for (TypedDependency td : tdls){// 依存关系的出发节点，依存关系，以及结束节点TreeGraphNode gov = td.gov();GrammaticalRelation reln = td.reln();String shortName = reln.getShortName();TreeGraphNode dep = td.dep();if (gov == rootNode){switch (shortName){case "nsubjpass":case "dobj":case "attr":mainPart.object = dep;break;case "nsubj":case "top":mainPart.subject = dep;break;}}if (mainPart.object != null && mainPart.subject != null){break;}}// 尝试合并主语和谓语中的名词性短语combineNN(tdls, mainPart.subject);combineNN(tdls, mainPart.object);if (!mainPart.isDone()) mainPart.done();}return mainPart;}private static MainPart getNPPhraseMainPart(Tree tree){MainPart mainPart = new MainPart();StringBuilder sbResult = new StringBuilder();List<String> phraseList = getPhraseList("NP", tree);for (String phrase : phraseList){sbResult.append(phrase);}mainPart.result = sbResult.toString();return mainPart;}//从句子中提取最小粒度的短语public static List<String> getPhraseList(String type, String sentence){return getPhraseList(type, lp.apply(seg(sentence)));}private static List<String> getPhraseList(String type, Tree tree){List<String> phraseList = new LinkedList<String>();for (Tree subtree : tree){if(subtree.isPrePreTerminal() && subtree.label().value().equals(type)){StringBuilder sbResult = new StringBuilder();for (Tree leaf : subtree.getLeaves()){sbResult.append(leaf.value());}phraseList.add(sbResult.toString());}}return phraseList;}//合并名词性短语为一个节点private static void combineNN(Collection<TypedDependency> tdls, TreeGraphNode target){if (target == null) return;for (TypedDependency td : tdls){// 依存关系的出发节点，依存关系，以及结束节点TreeGraphNode gov = td.gov();GrammaticalRelation reln = td.reln();String shortName = reln.getShortName();TreeGraphNode dep = td.dep();if (gov == target){switch (shortName){case "nn":target.setValue(dep.toString("value") + target.value());return;}}}}private static TreeGraphNode getRootNode(Collection<TypedDependency> tdls){for (TypedDependency td : tdls){if (td.reln() == GrammaticalRelation.ROOT){return td.dep();}}return null;}//分词private static List<Word> seg(String sentence){//分词LOG.info("正在对短句进行分词：" + sentence);List<Word> wordList = new LinkedList<>();List<Term> terms = HanLP.segment(sentence);StringBuffer sbLogInfo = new StringBuffer();for (Term term : terms){Word word = new Word(term.word);wordList.add(word);sbLogInfo.append(word);sbLogInfo.append(' ');}LOG.info("分词结果为：" + sbLogInfo);return wordList;}public static MainPart getMainPart(String sentence, String delimiter){List<Word> wordList = new LinkedList<>();for (String word : sentence.split(delimiter)){wordList.add(new Word(word));}return getMainPart(wordList);}public static void main(String[] args){/* String[] testCaseArray = {"我一直很喜欢你","你被我喜欢","美丽又善良的你被卑微的我深深的喜欢着……","只有自信的程序员才能把握未来","主干识别可以提高检索系统的智能","这个项目的作者是hankcs","hankcs是一个无门无派的浪人","搜索hankcs可以找到我的博客","静安区体育局2013年部门决算情况说明","这类算法在有限的一段时间内终止",};for (String testCase : testCaseArray){MainPart mp = MainPartExtractor.getMainPart(testCase);System.out.printf("%s\t%s\n", testCase, mp);}*/mpTest();}public static void mpTest(){String[] testCaseArray = {"我一直很喜欢你","你被我喜欢","美丽又善良的你被卑微的我深深的喜欢着……","小米公司主要生产智能手机","他送给了我一份礼物","这类算法在有限的一段时间内终止","如果大海能够带走我的哀愁","天青色等烟雨，而我在等你","我昨天看见了一个非常可爱的小孩"};for (String testCase : testCaseArray) {MainPart mp = MainPartExtractor.getMainPart(testCase);System.out.printf("%s   %s   %s \n",GraphUtil.getNodeValue(mp.getSubject()),GraphUtil.getNodeValue(mp.getPredicate()),GraphUtil.getNodeValue(mp.getObject()));}}
}

四、动态构建知识图谱

新建一个NodeServiceImpl，其中实现两个关键方法parseAndBind和addNode 首先是根据句子中抽取的主语或宾语在neo4j中创建节点的方法，这里根据节点的name判断是否为已存在的节点，如果存在则直接返回，不存在则添加：

@Service
@AllArgsConstructor
public class NodeServiceImpl implements NodeService {private final NodeRepository nodeRepository;private final RelationRepository relationRepository;@Overridepublic Node save(Node node) {Node save = nodeRepository.save(node);return save;}@Overridepublic void bind(String name1, String name2, String relationName) {Node start = nodeRepository.findByName(name1);Node end = nodeRepository.findByName(name2);Relation relation =new Relation();relation.setStartNode(start);relation.setEndNode(end);relation.setRelation(relationName);relationRepository.save(relation);}private Node addNode(TreeGraphNode treeGraphNode){String nodeName = GraphUtil.getNodeValue(treeGraphNode);Node existNode = nodeRepository.findByName(nodeName);if (Objects.nonNull(existNode))return existNode;Node node =new Node();node.setName(nodeName);return nodeRepository.save(node);}@Overridepublic List<Relation> parseAndBind(String sentence) {MainPart mp = MainPartExtractor.getMainPart(sentence);TreeGraphNode subject = mp.getSubject();    //主语TreeGraphNode predicate = mp.getPredicate();//谓语TreeGraphNode object = mp.getObject();      //宾语if (Objects.isNull(subject) || Objects.isNull(object))return null;Node startNode = addNode(subject);Node endNode = addNode(object);String relationName = GraphUtil.getNodeValue(predicate);//关系词List<Relation> oldRelation = relationRepository.findRelation(startNode, endNode,relationName);if (!oldRelation.isEmpty())return oldRelation;Relation botRelation=new Relation();botRelation.setStartNode(startNode);botRelation.setEndNode(endNode);botRelation.setRelation(relationName);Relation relation = relationRepository.save(botRelation);return Arrays.asList(relation);}}