​
Input

  • output (str): The generated text.
  • expectedOutput (str): The reference text.

​
Output

  • Result (float): A similarity score, typically between 0 and 1.

​
Interpretation

  • 1: The texts are considered semantically identical.
  • 0: The texts have completely different meanings.
  • Robust to paraphrasing and synonymous language.

​
Formula

Similarity=A⋅B∥A∥×∥B∥\mathrm{Similarity} = \frac{A \cdot B}{\|A\| \times \|B\|} Where A and B are the embedding vectors.
This is a similarity metric. Higher scores (closer to 1) indicate greater semantic similarity.

​
How It Works

Compute vector embeddings for the generated and reference texts, then measure their similarity (commonly cosine similarity). This captures shared meaning beyond exact word matches.

​
Use Cases

  • Evaluating chatbots and conversational AI
  • Assessing the quality of abstractive summaries
  • Measuring relevance in search and retrieval
  • Paraphrase detection