各社区PageRank计算：Cypher语句类型不匹配错误求助

问题描述 #

执行用于计算各社区PageRank的Cypher语句时，触发错误：

Type mismatch: expected String but was Map (line 3, column 26 (offset: 96)) "CALL gds.pageRank.stream({"

原语句如下：

MATCH (i:exp_amd_cif)
WITH distinct i.louvain_community as community
CALL gds.pageRank.stream({
nodeQuery:"MATCH (i:exp_amd_cif) WHERE i.louvain_community = $community
RETURN id(i) as id",
relationshipQuery: "MATCH (acs:exp_amd_cif)-[:transfer_to]->(act:exp_amd_cif) 
RETURN id(acs) as source, id(act) as target",
parameters:{community:community}})
YIELD nodeId, score
WITH community, nodeId, score
ORDER BY score DESC
RETURN community, 
collect(gds.util.asNode(nodeId).name)[..5] as top_5_representatives

错误原因 #

当前使用的Neo4j GDS版本不支持直接传入配置Map作为gds.pageRank.stream的参数，该版本要求第一个参数为已投影的图名称（字符串），而非配置Map，因此触发类型不匹配错误。此外原语句的关系查询未过滤社区范围，会导致计算结果基于全图关系，不符合社区内PageRank的计算预期。

解决方案 #

方案1：适配GDS 2.x+ 直接投影语法 #

如果环境使用GDS 2.x及以上版本，可使用直接投影配置Map的语法，同时修正关系过滤逻辑：

MATCH (i:exp_amd_cif)
WITH distinct i.louvain_community as community
CALL gds.pageRank.stream({
  nodeProjection: {
    exp_amd_cif: {
      label: 'exp_amd_cif',
      filter: 'n.louvain_community = $community'
    }
  },
  relationshipProjection: {
    transfer_to: {
      type: 'transfer_to',
      orientation: 'NATURAL',
      filter: 'startNode(r).louvain_community = $community AND endNode(r).louvain_community = $community'
    }
  },
  parameters: {community: community}
}) YIELD nodeId, score
WITH community, nodeId, score
ORDER BY score DESC
RETURN community, 
       collect(gds.util.asNode(nodeId).name)[..5] as top_5_representatives

方案2：兼容GDS 1.x 图投影+调用方式 #

针对旧版GDS，需先为每个社区创建临时图投影，计算完成后再删除投影：

MATCH (i:exp_amd_cif)
WITH distinct i.louvain_community as community
// 创建当前社区的临时图投影
CALL gds.graph.project(
  'temp-community-' + toString(community),
  {exp_amd_cif: {label: 'exp_amd_cif', filter: 'n.louvain_community = $community'}},
  {transfer_to: {type: 'transfer_to', filter: 'startNode(r).louvain_community = $community AND endNode(r).louvain_community = $community'}},
  {parameters: {community: community}}
) YIELD graphName
// 计算社区内PageRank
CALL gds.pageRank.stream(graphName) YIELD nodeId, score
WITH community, nodeId, score, graphName
// 删除临时图释放资源
CALL gds.graph.drop(graphName)
WITH community, nodeId, score
ORDER BY score DESC
RETURN community, 
       collect(gds.util.asNode(nodeId).name)[..5] as top_5_representatives

内容的提问来源于stack exchange，提问作者Zayyed Akhmed Assidqie