Persisting Spacy Vector Representations in Neo4j

I have nodes with text as a property stored in a Neo4j database. However, when I read text and create Spacy objects out of them to compute similarity between them and one text entered by user in memory, it takes too much time to execute. I wonder what is the best way to solve this problem?

I know that GraphAware exists, but it only supports OpenNLP and Stanford coreNLP. Another way I would think of is storing Spacy objects as another property to this node, but I do not know about performance when the size of the property is huge.

Thank you!