NEO4J-链路预测算法03-优先链接算法（Preferential Attachment）应用场景简介

说明：使用neo4j算法库时需引入跟neo4j数据库对应的算法库插件或自定义算法库

1.简介

        优先链接时一种用于计算节点接近度的度量。基于他们的共享邻居

        它使用如下公式进行计算的：,其中N(x)是与节点x相邻的节点集，N(y)是与y节点相邻的节点集，PN(x,y)即为计算x和y节点的相邻节点数乘积

值0表示两个节点不接近，而较高的值表示节点相近。

2.使用场景

        进入抖音app，常常会弹出一个框，可能认识，上面显示显示由多少个共同朋友，即使用公共邻居算法。

        neo4j算法库中提供了algo.linkprediction.preferentialAttachment函数

3.源码解析

源码如下：

    public double preferentialAttachment(@Name("node1") Node node1, @Name("node2") Node node2,
                                       @Name(value = "config", defaultValue = "{}") Map<String, Object> config) {
        if (node1 == null || node2 == null) {
            throw new RuntimeException("Nodes must not be null");
        }

        ProcedureConfiguration configuration = ProcedureConfiguration.create(config);
        RelationshipType relationshipType = configuration.getRelationship();
        Direction direction = configuration.getDirection(Direction.BOTH);

        return degree(node1, relationshipType, direction) * degree(node2, relationshipType, direction);
    }

入参:node1,node2为两个节点数据，不能为空，config为条件参数，config中能识别relationshipQuery和direction,其中relationshipQuery为节点关系编码，direction为方向，默认为BOTH无方向，OUTING为node1->node2,INCOMING为node2->node1。 

        方法解析如下：

1.将传入的config参数放入ProcedureConfiguration的config中

ProcedureConfiguration configuration = ProcedureConfiguration.create(config);

2.获取ProcedureConfiguration类config中relationshipQuery参数（此参数为关系名称）

RelationshipType relationshipType = configuration.getRelationship();

3.获取config中direction参数，此为指定关系方向

Direction direction = configuration.getDirection(Direction.BOTH);

4.条件返回节点1的关系节点，并且过滤掉与节点2没有关系的节点(即取节点1和节点2同时有relationshipType关系的且方向为direction的节点)

Set<Node> neighbors = new NeighborsFinder(api).findCommonNeighbors(node1, node2, relationshipType, direction);

5.统计节点node1具有relationshipType关系，且方向为derection的节点数

degree(node1, relationshipType, direction)：具体代码如下

    private int degree(Node node, RelationshipType relationshipType, Direction direction) {
        return relationshipType == null ? node.getDegree(direction) : node.getDegree(relationshipType, direction);
    }

node.getDegree(relationshipType, direction)的源码如下：作用统计该节点关系为relationshipType，指向为direction的关系的数量

    @Override
    public int getDegree(RelationshipType relationshipType, Direction direction) {
        return (int) Iterables.count(getRelationships(relationshipType,direction));
    }

node.getDegree(direction)的源码如下：作用统计该节点关系数量

    @Override
    public int getDegree(Direction direction) {
        return (int) Iterables.count(getRelationships(direction));
    }

4.neo4j中公共邻居算法使用示例

1.初始化数据

MERGE (zhen:Person {name: "Zhen"})
MERGE (praveena:Person {name: "Praveena"})
MERGE (michael:Person {name: "Michael"})
MERGE (arya:Person {name: "Arya"})
MERGE (karin:Person {name: "Karin"})

MERGE (zhen)-[:FRIENDS]-(arya)
MERGE (zhen)-[:FRIENDS]-(praveena)
MERGE (praveena)-[:WORKS_WITH]-(karin)
MERGE (praveena)-[:FRIENDS]-(michael)
MERGE (michael)-[:WORKS_WITH]-(karin)
MERGE (arya)-[:FRIENDS]-(karin)

2.以下将返回 Michael 和 Karin 的 Preferential Attachment 分数

MATCH (p1:Person {name: 'Michael'})
MATCH (p2:Person {name: 'Karin'})
RETURN algo.linkprediction.preferentialAttachment(p1, p2) AS score

结果：6.0

3.我们还可以根据特定的关系类型计算一对节点的分数。以下将仅根据FRIENDS关系返回 Michael 和 Karin 的 Preferential Attachment 分数

MATCH (p1:Person {name: 'Michael'})
MATCH (p2:Person {name: 'Karin'})
RETURN algo.linkprediction.preferentialAttachment(p1, p2, {relationshipQuery: "FRIENDS"}) AS score

结果：1.0

上一篇：NEO4J-链路预测算法02-公共邻居算法（adamicAdar）应用场景简介（抖音朋友推荐）