# Is there an example which shows the benefit of graph embedding

**URL:** <https://community.neo4j.com/t/is-there-an-example-which-shows-the-benefit-of-graph-embedding/60427>\
**Category:** Neo4j Graph Platform\
**Tags:** migrated\
**Created:** [June 21, 2022, 6:14pm UTC](https://community.neo4j.com/t/is-there-an-example-which-shows-the-benefit-of-graph-embedding/60427 "2022-06-21T18:14:58Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![guydomran](https://avatars.discourse-cdn.com/v4/letter/g/a183cd/32.png) [@guydomran](https://community.neo4j.com/u/guydomran)\
**Post date:** [June 21, 2022, 6:14pm UTC](https://community.neo4j.com/t/is-there-an-example-which-shows-the-benefit-of-graph-embedding/60427/1 "2022-06-21T18:14:58Z")

</div>

Have seen neo4j authored documentation and youtube videos on graph embedding. Following are the queries:

1. Can we do ML algorithms using GDS without graph embeddings.

2. If yes, Is there a concrete example of the benefit of graph embedding.

3. What are the indicators on when we should go for going for embeddings, is it just trial and error or there is a kind of rule or recommendation.

---

<div class="post-metadata">

**Author:** ![michael.hunger](https://sea1.discourse-cdn.com/flex021/user_avatar/community.neo4j.com/michael.hunger/32/27377_2.png) [@michael.hunger](https://community.neo4j.com/u/michael.hunger)\
**Post date:** [June 22, 2022, 11:20pm UTC](https://community.neo4j.com/t/is-there-an-example-which-shows-the-benefit-of-graph-embedding/60427/2 "2022-06-22T23:20:49Z")

</div>

1. yes you can also create feature vectors from properties (via numeric values, one-hot-encoding, scaling, neighbours/relationships within a slice of max-degree etc)  
2. similar to a word embedding in a document it automatically computes the "semantic essence" of a node, taking it's property and its topological context (attributes of relevant neighbours) into account, it's similar to "tell me who your friends are and I tell you who you are"  
3. graph embeddings is less manual effort than creating feature vectors yourself, you can use them for training external models, for creating kNN similarity graphs, for node classification, link prediction and property regression in the graph.