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

sql 复制代码

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

sql 复制代码

MATCH (p:Person) RETURN p

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：

sql 复制代码

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：

sql 复制代码

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

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：

sql 复制代码

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

SpringBoot整合neo4j

一、依赖

xml 复制代码

<?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

yml 复制代码

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

java 复制代码

//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：

java 复制代码

@Service
@AllArgsConstructor
public class NodeServiceImpl implements NodeService {
    private final NodeRepository nodeRepository;
    private final RelationRepository relationRepository;


    @Override
    public Node save(Node node) {
        Node save = nodeRepository.save(node);
        return save;
    }
    @Override
    public 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);
    }
    @Override
    public 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);
    }

}

测试：启动java应用，输入以下地址

http://127.0.0.1:8088/parse?sentence=海拉又被称为死亡女神

http://127.0.0.1:8088/parse?sentence= 死亡女神捏碎了雷神之锤

http://127.0.0.1:8088/parse?sentence=雷神之锤属于索尔

在图数据库neo4j里面查询：

MATCH (p:Person) RETURN p