A generic benchmark is always difficult as it is not representing your domain.
Some general tips:

• put enough operations ( createNode, createRelationship ) in the transactions 1k - 5k - 20k are good fits
• use -server
• give Neo4j enough heap-memory for your use-case
• run several rounds in a single benchmark, probably discard the first one and measure the others (to accomodate for JIT and Neo4j warmup)

What are your machine specs?
You can also have a look at these benchmarks which are used for performance regression testing.

这可能是因为您示例化了GraphDatabaseService并创建了节点。简单地在示例化的GraphDatabaseService上创建节点将花费较少的时间。

我认为@MichaelHunger说到点子上了。如果你花一秒钟写一个节点，你就有办法写很多事务。
您看过批次插入吗？
http://docs.neo4j.org/chunked/stable/batchinsert.html

就在今天，一个新的基准测试发布了。目前，它有两个图形数据库，Neo4j和Memgraph。在博客文章中，有一个链接到GitHub仓库和方法论。也许它会对你有帮助。
博客文章位于https://memgraph.com/blog/memgraph-vs-neo4j-performance-benchmark-comparison。