Recall vs. Hits

[kimwongyuda]

I appreciate to your nice work.

However, I am curious of how you evaluate the model.

In the mbeir_retriever.py file, you use below code as recall.
However, I think this is Hits@k not Recall@k.
is this metric Hits@k? or Recall@k?

def compute_recall_at_k(relevant_docs, retrieved_indices, k):
    # Recall used by CLIP and BLIP codebase
    # Return 0 if there are no relevant documents
    if not relevant_docs:
        return 0.0

    # Get the set of indices for the top k retrieved documents
    top_k_retrieved_indices_set = set(retrieved_indices[:k])

    # Convert the relevant documents to a set
    relevant_docs_set = set(relevant_docs)

    # Check if there is an intersection between relevant docs and top k retrieved docs
    # If there is, we return 1, indicating successful retrieval; otherwise, we return 0
    if relevant_docs_set.intersection(top_k_retrieved_indices_set):
        return 1.0
    else:
        return 0.0

Thank you.