Does the Conditional Probability of the Question Mentioned in the Paper Refer to the Computation of the Cross-Entropy Loss Function for the Question?

[xingyuma618]

Dear Author,

I hope this email finds you well. I recently read your article titled "Unsupervised-Passage-Reranking", and I found it to be a very innovative and intriguing piece of work. I truly appreciate your contribution to this field.

While reviewing the code implementation, I encountered some confusion. My understanding is that you concatenate the {passage}, the prompt "Please write a question based on this passage", and the {question} to compute the cross-entropy loss of the question. This loss value is then used as the passage's score for ranking purposes. Could you please confirm if my understanding is correct?

However, I noticed that the article mentions the conditional probability of the question, which does not seem to be explicitly reflected in the code. This has left me a bit puzzled.

I would greatly appreciate it if you could provide some clarification on this matter. Looking forward to your reply.

Best regards,