# Node Similarity Algorithm (Jaccard) inconsistent scores

**URL:** <https://community.neo4j.com/t/node-similarity-algorithm-jaccard-inconsistent-scores/55787>\
**Category:** Graph Data Science / Graph Analytics\
**Created:** [May 5, 2022, 2:39am UTC](https://community.neo4j.com/t/node-similarity-algorithm-jaccard-inconsistent-scores/55787 "2022-05-05T02:39:25Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![hlow](https://sea1.discourse-cdn.com/flex021/user_avatar/community.neo4j.com/hlow/32/23436_2.png) [@hlow](https://community.neo4j.com/u/hlow)\
**Post date:** [May 5, 2022, 2:39am UTC](https://community.neo4j.com/t/node-similarity-algorithm-jaccard-inconsistent-scores/55787/1 "2022-05-05T02:39:25Z")

</div>

Hi,  
I have the following graph visualization and I don't understand why the scores are not the same.

Let's assume that all 3 nodes (A, B, C) have an outgoing HAS relationship with american and casual ONLY. It makes sense to me that nodes A and B would have a perfect score of 1. However, why is the similarity score from nodes C to B only 0.75?

When I ran the gds.nodeSimilarity algorithm, I made sure to only include the HAS relationship and the two node types in the screenshot (blue and green). I also used a topK: 5 and similarityCutoff: 0.2.

Thank you

 ![Screen Shot 2022-05-04 at 7.37.35 PM (3)](https://us1.discourse-cdn.com/flex021/uploads/neo4jcommunity/original/3X/f/2/f28d12e2854caa2f4a58f686ed9b5fc646edae74.png)

---

<div class="post-metadata">

**Author:** ![hlow](https://sea1.discourse-cdn.com/flex021/user_avatar/community.neo4j.com/hlow/32/23436_2.png) [@hlow](https://community.neo4j.com/u/hlow)\
**Post date:** [May 5, 2022, 9:52pm UTC](https://community.neo4j.com/t/node-similarity-algorithm-jaccard-inconsistent-scores/55787/2 "2022-05-05T21:52:53Z")

</div>

Found the issue.  
It's because I included the "mutate" step and the mutateProperty value was in the projected graph.

If I skip that step and go straight to Write, then the scores are what I expected.