The similarity score is calculated by taking the cosine similarity between the embeddings of the concepts.