Can the ranking-filter successfully choose all the right schema items?

[Linhaowen2]

Hello, I am wondering when your top-Key_1 = 4 and top-Key_2 =5, can all of the schema items labels be chosen by the filter? In other words, would the case that the ranking-filter excludes the right schema items happens when you test or evaluate this model? If in that case, could you please tell me the percentage of correctness, in this ranking-filter part?

[lihaoyang-ruc]

Q: In other words, would the case that the ranking-filter excludes the right schema items happens when you test or evaluate this model?
A: Certainly, in the ranking stage, some correct schema items may indeed be excluded due to the inherent limitations of the neural network classifier. It's important to note that achieving 100% accuracy with a neural network is not feasible.

Q: If in that case, could you please tell me the percentage of correctness, in this ranking-filter part?
A: We assess the "recall" metric (or coverage) on the Spider dataset (with topk_1=4, topk_2=5) and have found the results to be satisfactory. Both table and column retrieval exhibit over 99% recall rates, indicating that only a minimal fraction of the correct schema items are inadvertently excluded.