Question about evaluation

[laituan245]

Hi. I have a question regarding the mention recall scores reported in your paper.

Table 2 says that the recall score of [Lee et al. (2018)] model is 96.6 when the mention/token ratio is 0.4.
However, as mentioned in various papers such as Lee et al. (2017) or Wu et al. (2020), it seems like the model's recall score is only around 92~93 when the ratio is 0.4.

Do you know why is there such as difference?